PE-Assembler: De novo assembler using short paired-end reads

Pramila Nuwantha Ariyaratne

Outline

• Method– Read screening– Seed building– Contig extension– Scaffolding– Gap filling

• Result

Data-sets Used

• Single end reads

• Paired end reads– ReadLength (from 25bp to 100bp)– Insert size vary from MinSpan to MaxSpan– The information are mainly from this data-sets.

Overview

• Read screening step select a set of reads as starting point.

• Seed building step extend these reads using Single End Reads to make them longer than MaxSpan. Successfully extended regions are called seeds.

• Contig extension try to extend all seeds using paired-end reads, result sequences called contigs.

Read screening

• Get all k-mers from all the reads.– A k-mer that is expected to occur in the actual

genome is called a ‘solid’ k-mer.– A k-mer that is expected to occur within a repeat

region is called a ‘repeat’ k-mer.• Repeat Region:– ACTTTGACACACACACAC……ACACACACGTTGAG

Read screening

Read screening

• A read is solid read if:– All it’s k-mers are within the two threshold cut-off.

• Example:– Two cut-off [42, 120] from previous graph.– K=5– Read: ACCGTATA– ACCGT, CCGTA, CGTAT, GTATA– 100, 70, 90, 140– Not a solid read.

Read screening

• Example:– Two cut-off [42, 120] from previous graph.– K=5– Read: ACCGTATG– ACCGT, CCGTA, CGTAT, GTATG– 100, 70, 90, 70– A solid read.

Seed Building

• Try to extend the solid read using all overlapping reads.

Seed Building• Because of sequencing errors or small repeats, there maybe

multiple feasible candidates.

Seed Building

• Ambiguities due to sequencing errors, we extend every candidate base up to ReadLength.– If only one candidate path reach the full distance

ReadLength, then that path is assumed to be correct extension.

• If no path or more than one path found. Try other side.

Seed Building

• Finally, when the sequence reach MaxSpan, (called seed) do a verification.

• At least one paired-end reads overlaps with this seed within expected length [MinSpan, MaxSpan]

Contig Extension

• This step aims to extend each verified seed to form a longer contig using Paired-End reads.

• For multiple feasible candidates, may due to 3 reasons.– First, sequencing errors. – Second, short tandem repeat. Handling in Gap

Filing step.– Third, long repeat. Which longer than MaxSpan.

Scaffolding

• Find the correct ordering of the resulting set of contigs.

• Gao Song currently working on it.

Gap filling

• Gap filling step is to assemble the gap region between two adjacent contigs to form a longer contig.

Gap filling

Simulated data results.

• Result compare using:– Average Length of all contigs.– N50, N90 of contigs. Bigger better.– Coverage.– Large Misassembly: accuracy is much more

important than others.

Simulated data results.  E. Coli S. Pombe HG18 chr10

  200bp + 10kbp 200bp + 1kbp + 10kbp 200bp + 10kbp 200bp + 1kbp + 10kbp 200bp + 1kbp + 10kbp

  PA Allpaths2 Velvet PA Allpaths2 PA Allpaths2 Velvet PA Allpaths2 PA

Contig statistics                      

Contigs (>200bp) 23 31 37 6 44 53 190 193 31 164 3158

Average length (kb) 202.7 155.4 125.6 777.4 107.6 231.8 65.2 63.7 394.7 75.3 39.1

Maximum length (kb) 2109.4 732.7 1506.8 2492.6 593.7 3500.9 868.4 1062.8 3519.6 851.0 514.5

Contig N50 size (kb) 883.9 357.1 1413.7 2492.6 362.7 1499.4 236.6 602.2 1487.7 226.8 89.0

Contig N90 size (kb) 355.7 92.4 597.4 2146.0 83.2 210.0 65.7 148.6 507.6 76.4 24.3

Coverage 99.89% 99.83% 99.60% 100.00% 99.85% 97.56% 98.62% 98.95% 97.78% 98.60% 94.20%

                       

Evaluation                      

Large misassemblies 0 0 2 0 0 0 1 1 0 0 1

Segment maps 99.20% 99.27% 95.00% 99.68% 99.18% 96.02% 96.78% 93.38% 96.42% 96.83% 90.48%

                       

Performance1                      

Execution time (min) 12 215 8 21 227 76 853 26 101 734 1682

Memory usage (gb) 1.3 15.4 2 2.3 29.7 3.8 45 5.3 4.5 66 16

Thank you for attention.