425
Design of “RNA-seq” Experiments RabadanLab 06/29/2015 Albert Lee

Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Embed Size (px)

Citation preview

Page 1: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Design of “RNA-seq”

Experiments

RabadanLab

06/29/2015

Albert Lee

Page 2: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation

Page 3: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation

• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses are being/will be called for.

Page 4: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated

analyses are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)

Page 5: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated

analyses are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

Page 6: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated

analyses are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

Page 7: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses

are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes

Page 8: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated

analyses are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes

• edgeR & DESeq : Modified Fisher Exact test for NB distribution

Page 9: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses

are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes• edgeR & DESeq : Modified Fisher Exact test for NB distribution• Cuffdiff : z-score based on the log-transformed ratio of expression divided by the variance

of the transformed ratio

Page 10: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses

are being/will be called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes• edgeR & DESeq : Modified Fisher Exact test for NB distribution• Cuffdiff : z-score based on the log-transformed ratio of expression divided by the variance

of the transformed ratio

Page 11: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses are being/will be

called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes• edgeR & DESeq : Modified Fisher Exact test for NB distribution• Cuffdiff : z-score based on the log-transformed ratio of expression divided by the variance of the transformed ratio

• One-factor-at-a-time method is ineffective. •

Page 12: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses are being/will be

called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes• edgeR & DESeq : Modified Fisher Exact test for NB distribution• Cuffdiff : z-score based on the log-transformed ratio of expression divided by the variance of the transformed ratio

• One-factor-at-a-time method is ineffective. •

Page 13: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Motivation• RNAseq is getting cheaper (~$600 per sample) and more sophisticated analyses are being/will be

called for.• Multifactorial analysis (multiple treatments, multiple tissues , multiple time points)• Repeated measures (more than one samples from the same individual)

• Currently, the majority of RNAseq tools heavily focus on comparing two classes• edgeR & DESeq : Modified Fisher Exact test for NB distribution• Cuffdiff : z-score based on the log-transformed ratio of expression divided by the variance of the transformed ratio

• One-factor-at-a-time method is ineffective. • Multifactorial options in edgeR/DESeq/Limma-Voom may be difficult to understand in the beginning

Page 14: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Aim• Goal is to identify a subset of genes ranked by interestingness while

accounting for the structure of the experimental design

Page 15: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Aim• Goal is to identify a subset of genes ranked by interestingness while

accounting for the structure of the experimental design

Example questions

Page 16: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Aim• Goal is to identify a subset of genes ranked by interestingness while

accounting for the structure of the experimental design

• (Level 1) What are the genes that are differentially expressed in tumor vs normal?

Example questions

Page 17: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Aim• Goal is to identify a subset of genes ranked by interestingness while

accounting for the structure of the experimental design

• (Level 1) What are the genes that are differentially expressed in tumor vs normal?

• (Level 2) What are the genes that are differentially expressed in tumor vs normal while controlling for batch effects?

Example questions

Page 18: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Aim• Goal is to identify a subset of genes ranked by interestingness while

accounting for the structure of the experimental design

• (Level 1) What are the genes that are differentially expressed in tumor vs normal?

• (Level 2) What are the genes that are differentially expressed in tumor vs normal while controlling for batch effects?

• (Level 3) What are the genes that are differentially expressed only in testis and not in other tissues?

Example questions

Page 19: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Aim• Goal is to identify a subset of genes ranked by interestingness while

accounting for the structure of the experimental design

• (Level 1) What are the genes that are differentially expressed in tumor vs normal?

• (Level 2) What are the genes that are differentially expressed in tumor vs normal while controlling for batch effects?

• (Level 3) What are the genes that are differentially expressed only in testis and not in other tissues?

• (Level 3) What are the genes that are differentially expressed between time 3 and time2 in drugged samples while controlling for vehicle effects?

Example questions

Page 20: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…

Page 21: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

Page 22: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

Page 23: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

data generation problem

Page 24: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.

data generation problem

Page 25: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.

data generation problem

non-normality problem

Page 26: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).

data generation problem

non-normality problem

Page 27: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).

data generation problem

non-normality problem

small replicate size problem

Page 28: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.

data generation problem

non-normality problem

small replicate size problem

Page 29: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.

data generation problem

non-normality problem

small replicate size problem

normalization problem

Page 30: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

non-normality problem

small replicate size problem

normalization problem

Page 31: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

There are many problems in RNAseq, but…Let’s assume that:

1. All heavy biological & computational work(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 32: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Level 1

Page 33: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Level 1 “What are the genes that are differentially expressed in tumor vs normal?”

Page 34: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.

Page 35: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 36: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 37: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 38: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 39: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 40: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 41: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 42: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 43: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 44: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 45: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

1. Completely randomized design

• Completely randomized design:• the simplest experimental design. • With this design, experimental units are randomly assigned to

treatments.Control Drug

Page 46: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 47: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

The typical matrix you will see

Page 48: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

The typical matrix you will see

Page 49: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition AThe typical matrix you will see

Page 50: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition AThe typical matrix you will see

Page 51: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition AThe typical matrix you will see

“Condition” factor

Page 52: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition AThe typical matrix you will see

“Condition” factor

N Experimental units

Page 53: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition AThe typical matrix you will see

“Condition” factor

N Experimental units

Each experimental unit is represented as a vector of gene expressions

Page 54: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition AThe typical matrix you will see

“Condition” factor

N Experimental units

Each experimental unit is represented as a vector of gene expressions

Expressionmeasurement

Page 55: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition AThe typical matrix you will see

Page 56: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

Page 57: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

Page 58: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

“In designing a microarray experiment,

Page 59: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

“In designing a microarray experiment, or RNA-seq

Page 60: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

“In designing a microarray experiment, we should concentrate on getting it right for one gene.

or RNA-seq

Page 61: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

“In designing a microarray experiment, we should concentrate on getting it right for one gene. As the other 53, 999 data points are measured on subsamples of the experimental unit(!),

or RNA-seq

Page 62: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

“In designing a microarray experiment, we should concentrate on getting it right for one gene. As the other 53, 999 data points are measured on subsamples of the experimental unit(!), they have no bearing on constructing a good design.”

or RNA-seq

Page 63: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

One thing to keep in mind

“In designing a microarray experiment, we should concentrate on getting it right for one gene. As the other 53, 999 data points are measured on subsamples of the experimental unit(!), they have no bearing on constructing a good design.”

-- George Casella, “Statistical Design”, Springer 2008

or RNA-seq

Page 64: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition A

Page 65: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition A

Page 66: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition A

Page 67: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition A

Page 68: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition A

expression22

113433247

1226799

14522

Page 69: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Condition BCondition A

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

Page 70: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

factor

Experimental units

Page 71: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

factor

Experimental units

a level of the factor

Page 72: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

factor

Experimental units

a level of the factor

another level of the factor

Page 73: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can compute sample means and sample standard deviations for two groups

Page 74: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can compute sample means and sample standard deviations for two groups

Page 75: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

You can compute sample means and sample standard deviations for two groups

Page 76: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

You can compute sample means and sample standard deviations for two groups

Page 77: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

,

You can compute sample means and sample standard deviations for two groups

Page 78: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

,

You can compute sample means and sample standard deviations for two groups

Null Hypothesis : =

Page 79: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

,

You can compute sample means and sample standard deviations for two groups

Null Hypothesis : =

Alternative Hypothesis :

Page 80: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

,

You can compute sample means and sample standard deviations for two groups

Null Hypothesis : =

Alternative Hypothesis :

+=Compute test-statistic :

Page 81: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

expression22

113433247

1226799

14522

conditionA

AAAABBBBB

,

,

You can compute sample means and sample standard deviations for two groups

Null Hypothesis : =

Alternative Hypothesis :

+=Compute test-statistic :

Compare with critical value & compute p value

Page 82: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Image source: http://www.beingencouraged.com/

Page 83: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 84: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 85: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 86: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Instead of focusing comparison of two groups,

Page 87: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 88: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 89: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 90: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Focus on modeling the expression as a function of factors

Page 91: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Page 92: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Expression

Page 93: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Expression

A Effect

Page 94: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Expression

A Effect B Effect

Page 95: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Expression

A Effect B Effect Error

Page 96: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

conditionA

AAAABBBBB

You can describe the same thing with linear modeling

Expression

A Effect B Effect Error

Expression = default effect + B Effect + Error

Page 97: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression(Y)22

113433247

1226799

14522

+ B+ B+ B+ B+ B

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

Use dummy variables to indicate a membership

Page 98: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression(Y)22

113433247

1226799

14522

Expression = default effect + B Effect + Error

+ B+ B+ B+ B+ B

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

Use dummy variables to indicate a membership

Page 99: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression(Y)22

113433247

1226799

14522

Expression = default effect + B Effect + Error

+ B+ B+ B+ B+ B

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

Use dummy variables to indicate a membership

[ | ] = +~ (0, )

Assume

Page 100: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression(Y)22

113433247

1226799

14522

Expression = default effect + B Effect + Error

+ B+ B+ B+ B+ B

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

Use dummy variables to indicate a membership

[ | ] = +~ (0, )

Assume

Estimate parameters, ,

Page 101: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression(Y)22

113433247

1226799

14522

Expression = default effect + B Effect + Error

+ B+ B+ B+ B+ B

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

Use dummy variables to indicate a membership

In R: lm ( expression ~ condition, data=data)

[ | ] = +~ (0, )

Assume

Estimate parameters, ,

Page 102: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

=

“Design matrix”“Response variable”

Page 103: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

=

“Design matrix”“Response variable”

Page 104: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

=

“Design matrix”“Response variable”

Page 105: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

=

“Design matrix”“Response variable”

Page 106: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

=

“Design matrix”“Response variable” “Coefficients”

Page 107: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1

=

“Design matrix”“Response variable” “Coefficients”

Page 108: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1S

=

“Design matrix”“Response variable” “Coefficients”

Page 109: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1S

=

“Design matrix”“Response variable” “Coefficients”

Page 110: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1S

=

=“Design matrix”“Response variable” “Coefficients”

Page 111: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1S

=

=“Design matrix”“Response variable” “Coefficients”

Page 112: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1S

=

=“Design matrix”“Response variable” “Coefficients”

Page 113: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression22

113433247

1226799

14522

Effect of B0

000011111

Default(A)

1

1

1

1

1

1

1

1

1

1S

=

=

Solve for

“Design matrix”“Response variable” “Coefficients”

Page 114: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

S

=

Solve for

Page 115: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

S

=

Solve for

Page 116: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

S

==

Solve for

Page 117: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

S

=== ( )

Solve for

Page 118: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

S

=== ( )=

Solve for

Page 119: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Expr

essi

on

Page 120: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Expr

essi

on

Page 121: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Expr

essi

on

Page 122: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Expr

essi

on

Page 123: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Expression = + +

Expr

essi

on

Page 124: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Expression = + +

Expr

essi

on

Page 125: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Expression = + +

Expr

essi

on

Page 126: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Expression = + +

Expr

essi

on

Page 127: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Drug effect

Expression = + +

Expr

essi

on

Page 128: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Drug effect

Expression = + +

Page 129: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Baseline effect

Drug effect

Expression = + +

Page 130: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

Page 131: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

Page 132: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

Page 133: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Page 134: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Now we can do analyze each term separately

Page 135: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Now we can do analyze each term separately

For example, we can check if drug effect is significant

Page 136: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

: = 0

Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Now we can do analyze each term separately

For example, we can check if drug effect is significant

Page 137: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

: = 0: 0

Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Now we can do analyze each term separately

For example, we can check if drug effect is significant

Page 138: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

: = 0: 0

~Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Now we can do analyze each term separately

For example, we can check if drug effect is significant

Page 139: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = + +

Baseline effect

Drug effect

: = 0: 0

~Here, we partitioned thevariance of expression into three parts:Baseline, drug effect, and residuals

Now we can do analyze each term separately

For example, we can check if drug effect is significant

Page 140: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 141: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

= Least square

Page 142: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )

Least square

Page 143: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )=

Least square

Since ( ) =

Page 144: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )=

=

Least square

Since ( ) =

Since Var(Y) =

Page 145: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )=

=

Least square

Since ( ) =

Since Var(Y) =

Since = and ==

Page 146: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )=

=

=

Least square

Since ( ) =

Since Var(Y) =

Since = and ==

Since =

Page 147: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )=

=

=

=

Least square

Since ( ) =

Since Var(Y) =

Since = and ==

Since =

Page 148: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

=

( ) = ( )=

=

=

=

= =

Least square

Since ( ) =

Since Var(Y) =

Since = and ==

Since =

Page 149: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 150: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

“Are the two means different?”

Page 151: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

“Are the two means different?” =+

Page 152: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

“Are the two means different?” =+

Page 153: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

“Are the two means different?”

“Is there a additional effect of B (if we have A as a baseline)?”

=+

Page 154: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

“Are the two means different?”

“Is there a additional effect of B (if we have A as a baseline)?”

=+

= +

Page 155: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Limitations of CRD

• If experimental units are heterogeneous, there might be high false positives or false negatives

Page 156: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Level 2 “What are the genes that are differentially expressed in tumor vs normal while controlling for batch effects?”

Page 157: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 158: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 159: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 160: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 161: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 162: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 163: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 164: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 165: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 166: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 167: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 168: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 169: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 170: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design Control Drug

Batch 1

Batch 2

Batch 3

Page 171: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design

expression22

113433247

1226799

145223221

conditionA

AAAAABBBBBB

batch1

122331

12233

1st factor 2nd factor

Page 172: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design

expression22

113433247

1226799

145223221

conditionA

AAAAABBBBBB

batch1

122331

12233

1st factor 2nd factor

Page 173: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design

expression22

113433247

1226799

145223221

conditionA

AAAAABBBBBB

batch1

122331

12233

1st factor 2nd factor

~ +

Page 174: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design

expression22

113433247

1226799

145223221

conditionA

AAAAABBBBBB

batch1

122331

12233

1st factor 2nd factor

~ +

Page 175: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

2. Randomized block design

Page 176: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 177: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 178: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Page 179: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

BaselineControl & batch 1

Page 180: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

BaselineControl & batch 1

Effect of Drug

Page 181: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2BaselineControl & batch 1

Effect of Drug

Page 182: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 183: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 184: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 185: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Effect of drug + batch 2 effect

Page 186: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Effect of drug + batch 2 effect

Page 187: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Effect of drug + batch 2 effect

Page 188: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Effect of drug + batch 2 effect

Page 189: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expr

essio

n

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Effect of drug + batch 2 effect

Effect of drug +batch 3 effect

Page 190: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 191: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 192: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 193: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 194: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 195: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 196: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 197: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

Page 198: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

We effectively have partitioned the variance.

Page 199: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

We effectively have partitioned the variance.That is, we separated batch effects from treatment

Page 200: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

We effectively have partitioned the variance.

Now we can test if there is a effect of drug and if it’s significantThat is, we separated batch effects from treatment

Page 201: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0

Now we can test if there is a effect of drug and if it’s significantThat is, we separated batch effects from treatment

Page 202: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of batch 2

Effect of batch 3

BaselineControl & batch 1

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0 Test statistics: ~

Now we can test if there is a effect of drug and if it’s significantThat is, we separated batch effects from treatment

Page 203: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0 Test statistics: ~

Now we can test if there is a effect of drug and if it’s significantThat is, we separated batch effects from treatment

Page 204: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0 Test statistics: ~

Now we can test if there is a effect of drug and if it’s significant

33.6132

That is, we separated batch effects from treatment

Page 205: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0 Test statistics: ~

Now we can test if there is a effect of drug and if it’s significant

( )0.710233.6132

That is, we separated batch effects from treatment

Page 206: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0 Test statistics: ~

Now we can test if there is a effect of drug and if it’s significant

t-statistic

47.33( )

0.710233.6132

That is, we separated batch effects from treatment

Page 207: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Effect of Drug

We effectively have partitioned the variance.

Null hypothesis : : = 0 Test statistics: ~

Now we can test if there is a effect of drug and if it’s significant

t-statistic

47.33

p-value

<2e-16( )

0.710233.6132

That is, we separated batch effects from treatment

Page 208: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Level 3 “What are the genes that are differentially expressed between drugged mice in time 2 and in time1 while controlling for vehicle effects?”

Page 209: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Research Question

• Setting: Mice are randomly assigned to drug and vehicle groups and their expression measures are obtained at time 1, time 2, and time3 (not repeated measure design)

Page 210: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 211: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 212: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 213: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 214: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 215: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 216: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 217: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 218: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 219: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 220: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 221: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 222: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

3. Time seriesControl Drug

Time 1

Time 2

Time 3

Page 223: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Data and model

Page 224: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Data and model

Page 225: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Data and model

Page 226: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Data and model

Model :

Page 227: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

~ +

Data and model

Model :

Page 228: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

~ + + :

Data and model

Model :

Page 229: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

~ + + :

Data and model

Interaction term

Model :

Page 230: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

~ + + :

Data and model

Interaction term

Interaction : the effect of a factor on the response variable isdifferent depending on the level of another factor

Model :

Page 231: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

~ + + :

Data and model

Interaction term

Interaction : the effect of a factor on the response variable isdifferent depending on the level of another factor

Model :

Ex) drug will act differently in t2 from control in t2

Page 232: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

In R..

Page 233: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

In R..

Page 234: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 235: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 236: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 237: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Page 238: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Page 239: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Page 240: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Page 241: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Page 242: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Page 243: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Page 244: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Time 2effect

Page 245: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Drug & time2 interaction

Time 2effect

Page 246: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Drug & time2 interaction

Time 2effect

Page 247: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Drug & time2 interaction

Time 2effect

Time3 effect

Page 248: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Drug & time2 interaction

Time 2effect

Time3 effect

Page 249: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Drug & time2 interaction

Drug & time3interaction

Time 2effect

Time3 effect

Page 250: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Time3 effect

Drug effect

Drug & time2 interaction

Drug & time3interaction

Time 2effect

Time3 effect

How do we interpretInteraction terms?

Page 251: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Drug & time2 interaction

Time 2effect

This segment corresponds tothe difference betweendrug_t2 and drug_t1

Page 252: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 253: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 254: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Because

Page 255: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

Page 256: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

Page 257: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

Page 258: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

minus

Page 259: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

minus

that

Page 260: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

minus

that

Page 261: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

minus

that

Page 262: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

minus

that equals

Page 263: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Becausethis

minus

that equals this

Page 264: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Drug & time2 interaction

Time 2effect

This segment corresponds todrug_t2 – drug_t1

Page 265: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Page 266: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

Page 267: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

Page 268: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to

Page 269: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to

Page 270: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to the difference between ctrl_t2 and ctrl_t1

Page 271: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to the difference between ctrl_t2 and ctrl_t1

Page 272: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to the difference between ctrl_t2 and ctrl_t1

Page 273: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to the difference between ctrl_t2 and ctrl_t1

Page 274: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Baseline effects

Time 2effect

By the same token,

this corresponds to the difference between ctrl_t2 and ctrl_t1

Page 275: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Drug & time2 interaction

Time 2effect

Time 2effect Therefore,

the interaction corresponds to

(drug_t2 – drug_t1) –(ctrL_t2 – ctrl_t1)

Page 276: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

Interaction: Difference of Differences

Page 277: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

Interaction: Difference of Differences

Page 278: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

(drug_t2 – drug_t1) (ctrL_t2 – ctrl_t1)

Interaction: Difference of Differences

Page 279: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

minus

(drug_t2 – drug_t1) (ctrL_t2 – ctrl_t1)-

Interaction: Difference of Differences

Page 280: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

minus =

(drug_t2 – drug_t1) (ctrL_t2 – ctrl_t1)-

Interaction: Difference of Differences

Page 281: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

minus

Drug & time2 interaction

=

(drug_t2 – drug_t1) (ctrL_t2 – ctrl_t1)-

Interaction: Difference of Differences

Page 282: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Drug & time2 interaction

Time 2effect

Time 2effect

minus

Drug & time2 interaction

=

(drug_t2 – drug_t1) (ctrL_t2 – ctrl_t1)-

Interaction: Difference of Differences

Does this gene respond differently to drug at time t2 vs time1 than to placebo at time2 vs time1?

Page 283: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Drug & time3interaction

Time3 effect

What about this one?

Time3 effect

Page 284: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

Page 285: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 286: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 287: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

drug:time2:

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 288: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

(drug_t2 – drug_t1) - (ctrl_t2 – ctrl_t1) drug:time2:

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 289: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

drug:time3(drug_t2 – drug_t1) - (ctrl_t2 – ctrl_t1) drug:time2:

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 290: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

drug:time3(drug_t2 – drug_t1) - (ctrl_t2 – ctrl_t1) drug:time2: (drug_t3 – drug_t1) - (ctrl_t3 – ctrl_t1)

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 291: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

drug:time3(drug_t2 – drug_t1) - (ctrl_t2 – ctrl_t1) drug:time2: (drug_t3 – drug_t1) - (ctrl_t3 – ctrl_t1)

So,

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 292: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

drug:time3(drug_t2 – drug_t1) - (ctrl_t2 – ctrl_t1) drug:time2: (drug_t3 – drug_t1) - (ctrl_t3 – ctrl_t1)

So, drug:time3 - drug:time2:

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 293: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

What if we want to compute

We know that drug:time2 corresponds to?

drug:time3(drug_t2 – drug_t1) - (ctrl_t2 – ctrl_t1) drug:time2: (drug_t3 – drug_t1) - (ctrl_t3 – ctrl_t1)

So, drug:time3 - drug:time2:

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2)

(drug_t3 – drug_t2) - (ctrl_t3 – ctrl_t2) ?

“What are the genes that are differentially expressed between drugged mice in time3 and in time2 while controlling for vehicle effects?”

Page 294: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 295: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 296: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

InterceptDrugTime2Time3Drug:time2Drug:time3

coefficients

Page 297: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

[0, 0, 0, 0, -1, 1] InterceptDrugTime2Time3Drug:time2Drug:time3

contrast coefficients

Page 298: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

[0, 0, 0, 0, -1, 1] InterceptDrugTime2Time3Drug:time2Drug:time3

=

contrast coefficients

Page 299: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

[0, 0, 0, 0, -1, 1] InterceptDrugTime2Time3Drug:time2Drug:time3

= Drug:time3 - Drug:time2

contrast coefficients

Page 300: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

[0, 0, 0, 0, -1, 1] InterceptDrugTime2Time3Drug:time2Drug:time3

= Drug:time3 - Drug:time2

contrast coefficients result

Page 301: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 302: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 303: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 304: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 305: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 306: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

general linear hypothesis test

Page 307: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

general linear hypothesis test

Page 308: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Use contrast matrix for arbitrary comparison

Expression = Intercept + drug+ time2 + time3+ drug:time2+ drug:time3

Page 309: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

If using interaction is confusing…

Page 310: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

If using interaction is confusing…

Make combined factor

Page 311: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

If using interaction is confusing…

Make combined factor

Page 312: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

If the interaction is confusing…

Page 313: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Fit classtime instead, and use contrast

Page 314: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Fit classtime instead, and use contrast

lm ( expression ~ classtime ,data = data)

Page 315: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Fit classtime instead, and use contrast

lm ( expression ~ classtime ,data = data)

Page 316: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

Page 317: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

nxhExpression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

Page 318: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

Page 319: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

Page 320: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Page 321: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast:

Page 322: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Expression = control_t1 + control_t2+ control_t3 + drug_t1 + drug_t2 + drug_t3

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 323: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 324: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 325: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 326: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 327: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 328: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Page 329: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Previous method

Page 330: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

To compute , (drug_t3- drug_t2) – (control_t3 – control_t2)

Use contrast: [ 0, 1, -1 , 0, -1, 1 ]

Previous method

Page 331: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Level 4 “Pre-Post-Control” Design

Page 332: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Study design

• Pre-post-control study design

control

Drug

pre treatment post treatment

Page 333: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Study design

• Pre-post-control study design

control

Drug

pre treatment post treatment

Page 334: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Study design

• Pre-post-control study design

control

Drug

pre treatment post treatment

Page 335: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

Page 336: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

Page 337: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• Using Interaction

Page 338: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables

Page 339: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables~ + : + :

Page 340: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables~ + : + :

Page 341: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables~ + : + :

Page 342: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables~ + : + :

Page 343: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables• Using Mixed model

Page 344: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables• Using Mixed model

• Use expression as dependent variable and use treatment and treatment and time interaction as fixed effects and patients as random effects.

Page 345: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• + : 1+ 1+ 1+)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables• Using Mixed model

• Use expression as dependent variable and use treatment and treatment and time interaction as fixed effects and patients as random effects.

~ + : + 1 + )

Page 346: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• + : 1+ 1+ 1+)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables• Using Mixed model

• Use expression as dependent variable and use treatment and treatment and time interaction as fixed effects and patients as random effects.

~ + : + 1 + )

Page 347: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• + : 1+ 1+ 1+)

• ~ + : + :

• Using Interaction • Use expression as dependent variable and use treatment, Patient and treatment interaction, and time

and treatment interaction as independent variables• Using Mixed model

• Use expression as dependent variable and use treatment and treatment and time interaction as fixed effects and patients as random effects.

~ + : + 1 + )

Assume Patients are random samples (from the hypothetical population)

Page 348: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

~

~ +

Page 349: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

~

~ +

Page 350: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• Using Gain of score

~

~ +

Page 351: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• Using Gain of score• Compute the difference between post and pre expression as a response variable

(Diff Expression) and use treatment as a independent variable~

~ +

Page 352: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 353: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 354: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 355: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 356: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 357: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 358: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

~

~ +

Page 359: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

• Using ANCOVA

~ +

Page 360: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

• Using ANCOVA • Use post as a response variable (Y) and fit the linear/general linear model

using factors and also pre as a covariate

~ +

Page 361: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• +

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

• Using ANCOVA • Use post as a response variable (Y) and fit the linear/general linear model

using factors and also pre as a covariate~ +~ +

Page 362: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• +

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

• Using ANCOVA • Use post as a response variable (Y) and fit the linear/general linear model

using factors and also pre as a covariate~ +~ +

Page 363: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

TIMTOADY(There’s more than one way to do it)

• +

• ~ • Using Gain of score

• Compute the difference between post and pre expression as a response variable (Diff Expression) and use treatment as a independent variable

• Using ANCOVA • Use post as a response variable (Y) and fit the linear/general linear model

using factors and also pre as a covariate~ +~ +

Page 364: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

In the real world…

Page 365: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

Page 366: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

Page 367: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

Page 368: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

data generation problem

Page 369: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.

data generation problem

Page 370: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.

data generation problem

non-normality problem

Page 371: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).

data generation problem

non-normality problem

Page 372: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).

data generation problem

non-normality problem

small replicate size problem

Page 373: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).

data generation problem

non-normality problem

small replicate size problem

Page 374: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.

data generation problem

non-normality problem

small replicate size problem

Page 375: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.

data generation problem

non-normality problem

small replicate size problem

normalization problem

Page 376: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

non-normality problem

small replicate size problem

normalization problem

Page 377: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 378: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Page 379: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Page 380: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Gee. Tophat takes hours and hours

Page 381: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Gee. Tophat takes hours and hours

Remedy:

Page 382: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Gee. Tophat takes hours and hours

Remedy:1) Use STAR aligner instead. It should take 4 to 8 minutes.

Page 383: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Gee. Tophat takes hours and hours

Remedy:1) Use STAR aligner instead. It should take 4 to 8 minutes. 2) Utilize multithreads.

Page 384: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Gee. Tophat takes hours and hours

Remedy:1) Use STAR aligner instead. It should take 4 to 8 minutes. 2) Utilize multithreads.

Ex) What could takes 30 minutes can be done in 30 seconds

Page 385: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Page 386: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

Page 387: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

“Not really. RNAseq is usually measured in counts”

Page 388: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

“Not really. RNAseq is usually measured in counts”

Remedy:

Page 389: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

1) Assume negative binomial and fit generalized linear model

“Not really. RNAseq is usually measured in counts”

Remedy:

Page 390: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. e have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemD

1) Assume negative binomial and fit generalized linear model

2) Transform your counts to log and fit linear model

“Not really. RNAseq is usually measured in counts”

Remedy:

Page 391: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 392: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 393: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemNot really. RNAseq sample size is usually 2~ 3 per group

Page 394: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemNot really. RNAseq sample size is usually 2~ 3 per group

Remedy:

Page 395: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemNot really. RNAseq sample size is usually 2~ 3 per group

Remedy:Use some sort of shrinkage method to stabilize variance(borrowing information of all genes)

Page 396: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problemNot really. RNAseq sample size is usually 2~ 3 per group

Remedy:Use some sort of shrinkage method to stabilize variance(borrowing information of all genes)Use edgeR, DESeq, Voom

Page 397: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 398: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 399: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Remedy:

Page 400: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Remedy:There are many normalization schemes. Choose one.

Page 401: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 402: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Page 403: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Remedy:

Page 404: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Remedy:1) Use Cufflinks, RSEM, Stringtie, etc.

Page 405: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

We have assumed that

1. All heavy biological & computational works(sequencing, preprocessing, alignment, and counting ) have been done.

2. Expression measurement follows normal distribution.3. We have a large size of experimental units(+100 samples).4. Normalization is taken care of.5. Expression measurements are collected at gene level.

data generation problem

Isoform problem

non-normality problem

small replicate size problem

normalization problem

Remedy:1) Use Cufflinks, RSEM, Stringtie, etc. 2) Exon level counting

Page 406: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Big Picture

Page 407: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Prof. Peter Sims’ Talks

Albert Lee’s Talk

Alexander Lachmann’s Talk

Prof. Friedman’s Talk

Page 408: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Prof. Peter Sims’ Talks

Albert Lee’s Talk

Alexander Lachmann’s Talk

Prof. Friedman’s Talk

Page 409: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

RNA-seq

Prof. Peter Sims’ Talks

Albert Lee’s Talk

Alexander Lachmann’s Talk

Page 410: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads

RNA-seq

Albert Lee’s Talk

Alexander Lachmann’s Talk

Page 411: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads

RNA-seq

“How do you efficiently capture the high quality biological data with little cost yet high resolution?”

Albert Lee’s Talk

Alexander Lachmann’s Talk

Page 412: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads

RNA-seq

“How do you efficiently capture the high quality biological data with little cost yet high resolution?”

Albert Lee’s Talk

Alexander Lachmann’s Talk

Page 413: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list

RNA-seq

“How do you efficiently capture the high quality biological data with little cost yet high resolution?”

Alexander Lachmann’s Talk

Page 414: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list

RNA-seq

“How do you efficiently capture the high quality biological data with little cost yet high resolution?”

“What is the most interesting subset of genes within our experiment design?”

Alexander Lachmann’s Talk

Page 415: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list Gene list to knowledge

RNA-seq

“How do you efficiently capture the high quality biological data with little cost yet high resolution?”

“What is the most interesting subset of genes within our experiment design?”

Page 416: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list Gene list to knowledge

RNA-seq

“How do you efficiently capture the high quality biological data with little cost yet high resolution?”

“What is the most interesting subset of genes within our experiment design?”

“What’s the biological meaning of those genes that are affected?”

Page 417: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list Gene list to knowledge

RNA-seq

Biological question

Page 418: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list Gene list to knowledge

RNA-seq

Biological question

Page 419: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Cells to reads Reads to ranked gene list Gene list to knowledge

RNA-seq

Biological question

Page 420: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 421: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015
Page 422: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

On that note….

Page 423: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Sir Ronald Aylmer FisherFather of “Design of Experiments”

On that note….

Page 424: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

Sir Ronald Aylmer FisherFather of “Design of Experiments”

“The statistician cannot evade the responsibility for understanding the process he applies or recommends”

On that note….

Page 425: Design of “RNA-seq” Experiments - Columbia Universitysystemsbiology.columbia.edu/sites/default/files/Design_of_rnaseq... · Design of “RNA-seq” Experiments RabadanLab 06/29/2015

THE END