paired-end read length for de novo assembly

[Seqasaurus]

I've just been told by a sequencing company rep that 50-bp PE reads are "better" for de novo assembly of bacterial genomes than 100-bp PE reads. This seemed dubious to me. At this company a 50-bp PE lane is £1000 cheaper. Is there any truth in this?

[kmcarr]

For a bacterial de novo assembly I would be plenty happy to use 50 bp PE reads as opposed to 100 bp. The only thing 100bp reads will get you in this case is increased coverage. You should have no problem getting the coverage required for a bacterium using a single lane of 50 bp reads. In fact one lane will give you massively more coverage than you need. The 50 bp reads will also be (on average) more accurate than the 100 bp (quality/accuracy declines as the reads get longer) making the assembly task easier and higher confidence in the resultant consensus. Throw in the bonus of saving £1,000 and it's a no-brainer.

[pmiguel]

£1000? Is this for finished sequence, maybe? Seems like you could do 10 bacterial genomes with a lane of sequence -- at draft quality. Diminishing returns at more than 100x coverage, I think.

--
Phillip

[Seqasaurus]

thanks for the advice. Don't worry, I wasn't planning on wasting an entire lane on one 3-mb genome. There'll be a few bacterial genomes and an insect transcriptome on there too.

that's £4000 instead of £5000 (and no bioinformatics). Pretty much twice as expensive as a core facility over here...

[pmiguel]

By the way, I don't believe that 50 nt PE reads are better for bacterial genome assembly than 100 nt PE reads. We generally get our best assemblies (fewest/largest contigs) with ABySS-PE set to a kmer size of around 70. Since 50 nt reads would preclude using a kmer that long, my guess is that your assembly with 50 nt PE will not be as good as with 100 nt PE reads.

That said, we have not actually done the experiment where we chop the reads off at 50 nt and see what the contigs look like at the same coverage.

--
Phillip