Metrology & Predictive Design For Synthetic Biology Jacob Beal November, 2013

Metrology & Predictive DesignFor Synthetic BiologyJacob Beal

November, 2013

Bringing Wet & Dry Together…

Computer sciencetools and models

IPTG not green

LacI

AIPTG

BGFPoutputs outputs outputsarg0arg0

LacI A

IPTG

B GFP

O(t)P(I,t) +

Delay

I(t)

Dilution& decay

Deeper Understanding

Reliable Engineering

Outline

• Vision and Motivation• Proto BioCompiler• Calibrating Flow Cytometry• Building EQuIP Models• Prediction & Validation

Vision: WYSIWYG Synthetic Biology

Bioengineering should be like document preparation:

4

Why is this important?

• Breaking the complexity barrier:

• Multiplication of research impact• Reduction of barriers to entry

*Sampling of systems in publications with experimental circuits

1975 1980 1985 1990 1995 2000 2005 2010100

1,000

10,000

100,000

1,000,000

207

2,100 2,700

7,500 14,600

32,000

583,0001,080,000

Year

Leng

th in

bas

e pa

irs

DNA synthesis Circuit size ?

5

[Purnick & Weiss, ‘09]

Why a tool-chain?

Organism Level Description

Cells

This gap is too big to cross with a single method!

6

TASBE tool-chain


Abstract Genetic Regulatory Network

DNA Parts Sequence

Assembly Instructions

Cells

High level simulator

Coarse chemical simulator

Testing

High Level DescriptionIf detect explosives: emit signalIf signal > threshold: glow red

Detailed chemical simulator

Modular architecture also open for flexible choice of organisms, protocols, methods, …

7

RonWeiss

Douglas Densmore

Collaborators:

[Beal et al, ACS Syn. Bio. 2012]

A Tool-Chain Example

If detect explosives: emit signalIf signal > threshold: glow red

(def simple-sensor-actuator () (let ((x (test-sensor))) (debug x) (debug-2 (not x))))

Mammalian Target E. coli Target

8

A high-level program of a system that reacts depending on sensor output





9

Program instantiated for two target platforms





10

Abstract genetic regulatory networks





11

Automated part selection using database of known part behaviors





12

Automated assembly step selection for two different platform-specific assembly protocols





Uninduced Uninduced

Induced Induced

13

Resulting cells demonstrating expected behavior


Outline


Focus: BioCompiler


Abstract Genetic Regulatory Network

DNA Parts Sequence

Assembly Instructions

Cells

High level simulator

Coarse chemical simulator

Testing

High Level DescriptionIf detect explosives: emit signalIf signal > threshold: glow red

Detailed chemical simulator

Compilation &Optimization

15

Other tools aiming athigh-level design:Cello, Eugene, GEC,GenoCAD, etc.

[Beal, Lu, Weiss, 2011]

Transcriptional Logic Computations

16

Operators translated to motifs:

IPTG not green

LacI

AIPTG

B

GFPoutputs outputs outputsarg0arg0

LacI A

IPTG

B GFP

Motif-Based Compilation

17

(def sr-latch (s r) (letfed+ ((o boolean (not (or r o-bar))) (o-bar boolean (not (or s o)))) o))

(green (sr-latch (aTc) (IPTG)))

Design Optimization

LacI B

IPTG

I

G

IF

GFP

DE1 E2

A

aTcJ

HC

JTetR

Unoptimized: 15 functional units, 13 transcription factors 18

GFP

Design Optimization

LacIF

IPTG

TetR

H

aTc

F

Final Optimized:5 functional units4 transcription factors

(def sr-latch (s r) (letfed+ ((o boolean (not (or r o-bar))) (o-bar boolean (not (or s o)))) o))

(green (sr-latch (aTc) (IPTG)))

Unoptimized: 15 functional units, 13 transcription factors

H

19

Complex Example: 4-bit Counter

(4-bit-counter)

compiled for mammalian

Optimized compiler already outperforms human designers

Barriers & Emerging Solutions:

• Barrier: Availability of High-Gain Devices– Emerging Solution: combinatorial device libraries

based on TALs, ZFs, miRNAs

• Barrier: Characterization of Devices– Emerging solution: TASBE characterization method

• Barrier: Predictability of Biological Circuits– Emerging solution: EQuIP prediction method

Outline

• Vision and Motivation• Proto BioCompiler• Calibrating Flow Cytometry TASBE

Method• Building EQuIP Models• Prediction & Validation

[Beal et al., Technical Report: MIT-CSAIL-TR-2012-008, 2012]

First, some metrology…

Unit mismatch!

Output[R2]

[R1][O

utpu

t][R2]

R2R1Arbitrary Red

Arbi

trar

y Bl

ue

Arbitrary RedAr

bitr

ary

Blue

How Flow Cytometry Works

Challenges:• Autofluorescence• Variation in measurements• Spectral overlap• Time Contamination

• Lots of data points!• Different protein fluorescence• Individual cells behave (very)

differently

Fluorescent Beads Absolute Units

Run beads every time: flow cytometers drift up to 20 percent!

Also can detect instrument problems, mistakes in settings

SpheroTech RCP-30-5A

Compensating for Autofluorescence

Negative control used for this

Compensating for Spectral Overlap

Strong positive control used for each colorNote: only linear when autofluorescence subtracted

Translating Fluorescence to MEFL

• Only FITC channel (e.g. GFP) goes directly

• Others obtained from triple/dual constitutive controls• Must have exact same constitutive promoter! • Must have a FITC control protein!

Outline


[Davidsohn et al., IWBDA, 2013]

TASBE Characterization Method

Transient cotransfection of 5 plasmids

Calibrated flow cytometry

Analysis by copy-count subpopulations

OutputDox R1

pCAG

Dox

T2ArtTA3 VP16Gal4 pTREEBFP2

pTRER1

pUAS-Rep1EYFPpCAGmkate

pCAG

Multi-plasmid cotransfection!?!

• Avoids all problems with adjacency, plasmid size, sequence validations

• Variation appears to be independent

Result: Input/Output Relations

R1 = TAL14 R1 = TAL21

32

Transfer curve for TAL 14 Transfer curve for TAL 21

Expression Dynamics

Fraction Active Mean Expression

Results division rate, mean expression time, production scaling factor

EQuIP model

Model = first-order discrete-time approximation

Output(t)

Regulated Production

Input(t)

Loss

Input

ΔOut

put …

λ

Time

Outline


[Davidsohn et al., IWBDA, 2013]

EQuIP Prediction

pCAG

Dox

T2ArtTA3 VP16Gal4 pTREEBFP2

pTRER1

pUAS-Rep1 pUAS-Rep2EYFPR2pCAGmkate

pCAG

R2 OutputDox R1

+


Input(t)

Loss

Input

ΔOut

put

…

λ

Time Output(t)


Loss

Input

ΔOut

put

…

λ

Time

Incremental Discrete Simulation[TAL21] [TAL14] [OFP]

TAL14 OFP

TAL21state

TAL14production TAL14

state

OFPproduction OFP

statetime

hour 1

hour 2

hour 46

loss

production

production

loss

production +

-

+

+ +

+

+

-

…

- -

-

…

-

…

High Quality Cascade Predictions

TAL14 TAL21 TAL21 TAL14

Circles = EQuIP predictionsCrosses = Experimental Data

1.6x mean error on 1000x range!

Summary

Automation supports design and debugging of biological devices, sensors, actuators, circuits– BioCompiler automates regulatory network design– TASBE method calibrates flow cytometry data– Cotransfected test circuits give good models– EQuIP accurately predicts cascade behavior from

models of individual repressors

Looking forward:• Real measurements good engineering• Bigger, better circuits on more platforms

[multiple manuscripts in preparation]

Acknowledgements:

Aaron Adler

Joseph Loyall

Rick Schantz

Fusun Yaman

Ron Weiss

Jonathan Babb

Noah Davidsohn

Ting Lu

Douglas Densmore

Evan Appleton

Swapnil Bhatia

Traci Haddock

Chenkai Liu

Viktor Vasilev

40

Characterization Tools Online!

Register: individual accounts or group account?

Anonymous access also available (but not private)

• On first use, you will have to terms of service

• Your data is secure, and can’t be shared on site.

• FireFox recommended; Chrome has an image-display bug.

https://synbiotools.bbn.com/

https://synbiotools.bbn.com/

Documents

Metrology & Predictive Design For Synthetic Biology Jacob Beal November, 2013