Creating psuedo paired-end sequencing reads from single-end reads

[cburke04]

Does anyone know of a program out there that will generate psuedo paired-end sequencing reads from Illumina 100bp single-end reads? This is for an initial pass at looking for structural variation (large scale inversions) between a reference genome and some high coverage SE read population genomics data that I have, I would like to use Pindel and some other methods that required paired-end sequencing data.

[dpryan]

I'm not aware of such a program, but depending on exactly what you wanted it wouldn't be difficult to write. My assumption is that you just want to split the 100bp read into two non-overlapping 50bp reads, which is simple to write (just reverse complement the sequence for read #2 and reverse its quality line). That should be doable in basically any language, though either of biopython or bioperl have the advantage of providing reverse complement functions, which means the whole thing could be a small handful of lines.

[westerman]

You may also need to put a gap between the paired ends. Say bases 1-45 and then 55-100. While I would tend to do what dpryan suggest -- roll your own -- that is just because it would be easy to do. Another thought is to use the FastX tools.

fastx_trimmer to get the first 45 (or 50) bases
fastx_reverse_complement and then fastx_trimmer to get the last 45 (50) bases.

[Brian Bushnell]

I have a program which does that...

bbfakereads.sh in=reads.fastq out1=r1.fastq out2=r2.fastq length=100

That will generate fake pairs from the input file, with whatever length you want (maximum of input read length). We use it in some cases for generating a fake LMP library for scaffolding from a set of contigs. Read 1 will be from the left end, and read 2 will be reverse-complemented and from the right end; both will retain the correct original qualities. And " /1" " /2" will be suffixed after the read name.

[dpryan]

Man, BBMap really is a swiss army knife. Does it come with a tooth-pick too?

[Brian Bushnell]

It does do contaminant removal As long as whatever is stuck between your teeth is genetically distinct from human, at least.

[dcameron]

I would have thought using a soft-clip based SV caller (local alignment required) would be more appropriate for single-end reads. A mean fragment untemplated sequence length of 0 with a variance of 0 does seem rather unusual and I wouldn't be surprised if some of the callers you tried crashed or gave meaningless results (eg: on my idealised simulated indel data, breakdancer-max drops to ~2% TP calling rate, and clever & gasv-pro refuse to run when average fragment size < 2*read length). I'd be interested to hear how you go.