Assembling a mixed DNA sample

[rubi]

Hi all,
I sequenced (Illumina GAII, 75 bp paired end reads, average insert 250) genomic DNA from a sample which contains both a host's and its parasite's DNA, where it is enriched with the parasite's DNA.

I'm only interested in assembling the parasite's genome, and it is a non reference genome so I'm doing denovo assembly.

As a filtering step I removed any of the reads that mapped to the host's EST data (there's no reference genome foe the host either), but that probably only removed protein coding DNA.

So I'm still stuck with non-coding DNA from the host, which I would like to remove. My thought was that since the reads are enriched with the parasite's DNA, contigs originating from the host should have low coverage.

The problem is that the assembly, produced using SOAPdenovo, uses Kmers to produce contigs and the result is that 80% of the contigs are as 100 bp short and no read maps to them so they have 0 coverage.

So my question is how do I work around this problem?
Should I use a larger Kmer size? Should I assume that these short contigs indeed represent the host's non-coding DNA and just filter them?

Thanks

[FredericRaymond]

Hello,

Have you tried aligning the reads to the human genome to filter the reads that are from the host? Afterwards, you could counterverify the filtered reads by aligning them to related parasite genomes to make sure you do not miss some parasite sequences.

In addition, you may want to try the Ray assembler, which also uses K-mers and performs very well with paired data. The use of this assembler may even solve part of you problem, since Ray considers the paired-read context to created contigs from the k-mer-based graph.

FR

[HESmith]

If the host-parasite relationship is not obligate, you could sequence just the host genome, assembly it, and use it to filter your existing data. Read depth would serve as an independent validation of this strategy.

[pytheus]

If you're studying something with exteme's of GC content like Plasmodium falciparum (~20%) you might be able to discriminate on GC content.