filtering out human seqs from metagenomic reads

[ssully]

What's considered to be the best tool for this -- removing human sequences from large sets of metagenomic next-gen reads? We tried BMTagger at default values , on a set of ~200 million 100nt Illumina reads, and it left in a lot of reads that hit human seqs with high confidence in subsequent blastn search vs. the NCBI nt database.

[rhinoceros]

What about bowtie2 against the human genome? They even have prebuilt indexes available. Blastn of over 100M reads against nt sounds rather wasteful use of computing resources..

[ssully]

Yes, it's far from good. But that's how many were left in our metagenomic set after filtering out short reads, duplicate reads, and (via BMTagger) human reads. So we''d like to try a better human read remover, to help insure that the final read set for downstream analysis (e.g. blastn) is all nonhuman. And smaller.

[rhinoceros]

Yes, it's far from good. But that's how many were left in our metagenomic set after filtering out short reads, duplicate reads, and (via BMTagger) human reads. So we''d like to try a better human read remover, to help insure that the final read set for downstream analysis (e.g. blastn) is all nonhuman. And smaller.

If I were you, I'd do trimming, bowtie2 against the human genome, assembly, and then blasts. Although for certain things like species distribution, assembly tends to introduce rather big bias (in my experience it increases the apparent presence of the most common taxa).

p.s. If you have human reads, you probably have other contaminants too, like bacteria from human skin among other stuff. Keep that in mind especially if your contamination rate is high..

[ssully]

We don't want to do assembly, because our main goal is to interrogate the diversity of taxa in our samples. We've done quality score filtering, length filtering, adapter trimming, duplicate removal - more vigorous quality trimming may be detrimental to uncovering diversity according to this study

We are studying a surface microbiome that humans interact with, so we don't mind skin bacteria; we want to catalog those, as well as any eukaryotic seqs. We don't even 'mind' the human sequences, it's just that their numbers make the seq files very large, so we want to split them out and treat human/nonhuman sets separately.

[GenoMax]

Perhaps one of these would be useful:

http://edwards.sdsu.edu/labsite/index.php/robert/301-how-to-remove-human-dna-sequence-contamination-from-metagenomes

http://clovr.org/hmp-dacc/hmp-dacc-clovr-human-contaminant-screening-walkthrough/

[lac302]

deconseq?...i haven't used it for anything larger than microbial genomes, but it works fairly well.

[leopal]

Great help!

Perhaps one of these would be useful:

http://edwards.sdsu.edu/labsite/index.php/robert/301-how-to-remove-human-dna-sequence-contamination-from-metagenomes

http://clovr.org/hmp-dacc/hmp-dacc-c...g-walkthrough/

These links are quite good options!

Thanks!