Remove subset of good reads to improve genome assembly

[NYGen]

Hello seqanswers,

I'm de novo assembling a diploid genome (haploid genome = 200Mb) with reads from Illumina Hiseq V4 2x125 (high-output) from a full lane of one library only with an insert size = 500bp. I have a question in the interest of reducing the computational challenges of DBG-based assembly:

Using Velvet, my contig and scaffold N50 is ~3kbp and ~30kbp, respectively. I've BLASTed through my scaffolds and discovered that I have my mitochondrial genes well-represented in my assembly contigs. Is it possible to remove the reads that uniquely map to the mitochondrial genome's contigs from my unassembled reads? The goal would be to lessen the memory load, and lessen the set of edges in the DBG in order to make the edge likelihood-testing process more-robust.

Is this a feasible move? Is it likely to help my assembly?

Thanks very much for any insights,
Josh

[Brian Bushnell]

Hi Josh,

That is one of the goals of BBSplit, which works best in situations where you have two different references (or sets of contigs) that you want to separate; in your case, the mito contigs and the main genome contigs. It's generally best to separate the reads of mito and non-mito and assemble them independently (typically subsampling the mito reads) because the coverage is so different.

For this case you may want to run with more sensitive settings than BBSplit's defaults, something like
bbsplit.sh in=reads.fq ref=mito.fa,main.fa basename=out_%.fq outu=unmapped.fq ambig2=split minid=0.75