Using Taxometer on 200 metagenomes

[bhagavadgitadu22]

Thanks for the great tool!

I want to try using taxometer on 200 metagenomes from a similar environment (stream biofilms with about 200 million paired-end reads of 150bp per sample). I have 40 million contigs>2000bp (about 200,000 per sample) that I annotated with MMseqs2 (database GTDB).

Because they are contigs I cannot really cluster them together except for the complete organisms (viruses....). Indeed even if the same organism is in 2 samples I do not see any reason why it would get fragmented in the same locations.

If I do not cluster though, primary read mappings are gonna be diluted between similar samples and I fear it will affect significantly the abundances. Indeed if the same organism is present in 2 samples very similar DNA is gonna be repeated twice and the reads are gonna map randomly to one or the other: the abundance of each organism per sample would be divided by 2 and this problem would scale with the number of samples where this organism is present.

Do you think I can run Taxometer despite this problem? Or could I allow multimapping to partly solve it (if Taxometer do not only consider primary reads)?

Also is Taxometer gonna be able to process that many contigs on 1 GPU and in less than 3 days? Over that I would have a problem in terms of available resources ;)

Thanks,

[jakobnissen]

Taxometer can use either pycoverm to estimate abundance, or input the abundances from a precomputed TSV file. Pycoverm does not use multimapping, and so does indeed dilute out the reads. However, in experiments, we have found that adding multi-mapping reads has no effect. It appears that this dilution is not a problem, at least below one million contigs. However, we have not tested anywhere close to 40M contigs.
Another option is to use strobealign --aemb, which will also be faster. This uses multi-mapping, splitting the reads deterministically as opposed to randomly.
My advice is to give it a try with strobealign --aemb.

I'm not sure it can do 40M contigs in 4 days on a GPU. I would guess yes, but you'll have to give it a go. Good luck!

[jakobnissen]

I'm going to close this as addressed