de novo assembly of PE reads

[lkral]

I am about to have DNA sequenced on a HiSeq and I expect about a 30 fold coverage of a 1x10^9 bp genome with 100 bp PE reads. I am unsure of the size of the fragments I should use to get the best likely assembly from these PE reads. I am aware that the best results would be obtained by having PE reads from several libraries of varying sizes but I can only afford to sequence one library at this time. Currently I would hope to obtain contigs that at least average 2,000 to 10,000 bp so single genes would likely be within a contig. The most likely problem in assembling a contig that spans a gene would be STRs in introns. I was thinking that a 1000 bp library should span across most such STRs. Any suggestions would be appreciated.

[lh3]

From the SGA paper: SGA can assemble 35X human reads into 10kbp contigs with reads from a single library with an average ~400bp insert size. Don't go for >500bp insert size. If I am right, the throughput and the quality of Illumina sequencing will degrade significantly.

[nickloman]

I think the drop off in yield comes around 800bp fragment size (which depending on how you calculate insert size would agree with Heng's suggestion of 500bp + paired-end 150bp reads).

[lh3]

By insert size, I always mean external size (i.e. the largest possible calculation). It is interesting that Illumina can now do 800bp (without a price?). I did not know this.

[nickloman]

We routinely do 500-600 base fragments and it works well. I think I read on another thread that 800 bases is where performance falls off a cliff, not tested that high ourselves.

[kmcarr]

We routinely do 500-600 base fragments and it works well. I think I read on another thread that 800 bases is where performance falls off a cliff, not tested that high ourselves.

I have one example from ~4 months ago running 3 libraries prepared from the same DNA sample with varying insert sizes. The estimated insert sizes (from BioAnalyzer) were 405, 540 and 795 bp. (Those are the insert size, not including adapters.) These libraries were run on a HiSeq 2000, one library per lane, 2x100bp PE. I observed no significant difference in the overall read quality among these three libraries, but the quality for all lanes on this flow cell was somewhat lower than typical, particularly at the end of read2.

[lh3]

Thanks for the info. I need to update my old knowledges.

[lkral]

Thank you all. I'll play it safe and with a 500 to 600 bp library.

[mjp]

I am about to have DNA sequenced on a HiSeq and I expect about a 30 fold coverage of a 1x10^9 bp genome with 100 bp PE reads. I am unsure of the size of the fragments I should use to get the best likely assembly from these PE reads. I am aware that the best results would be obtained by having PE reads from several libraries of varying sizes but I can only afford to sequence one library at this time. Currently I would hope to obtain contigs that at least average 2,000 to 10,000 bp so single genes would likely be within a contig. The most likely problem in assembling a contig that spans a gene would be STRs in introns. I was thinking that a 1000 bp library should span across most such STRs. Any suggestions would be appreciated.

Depending on the complexity of your organism (ploidity, heterozygosity level etc.) you will most likely not avoid sequencing large insert size libraries (5kb, 10kb, 20kb) to get a reasonable assembly. There are numerous papers on de novo assembly (of generaly relatively large genomes) that continuously use large insert size libraries to obtain good assembly.

It's hard to say whether you should see contigs of the size you mentioned, since again it all depends of the complexity level of your organism.

[lh3]

My view is libraries with large insert size mainly helps scaffolding, but not much for contigs. For example, SGA assembles reads with ~400bp insert to 10kb. Allpaths-LG assembles reads from variety of insert sizes to ~20kb. The contig N50 is not that different especially given that allpaths-lg uses 3-fold as many data which are much higher in cost. The scaffold N50 of allpaths-lg is by far better.

[lkral]

The current project is phase I where all I need to do is obtain contigs that are large enough to contain a gene or part of a gene. If contigs are smaller than a gene I can align these to orthologs from other fish species for assembly of those genes. In phase II in about a year or so, I hope to build longer scaffolds aligning to long oxford nanopore generated sequences (I trust these nanopores will work as advertised).