Open ctb opened 2 years ago
per conversation in https://github.com/sourmash-bio/sourmash/issues/2266, I think adapter trimming and quality filtering is also generally unnecessary. Although if QC programs like FastQC tell you you have a major problem, that might be something worth doing (or, resequencing ;).
here are some results from sourmash gather running on four different metagenomes after the report-both-weighted-and-unweighted PR was merged, https://github.com/sourmash-bio/sourmash/pull/2301:
(duplicated from https://github.com/dib-lab/genome-grist/issues/197#issuecomment-1263687946)
here are the results!
metagenome | unweighted match | weighted match |
---|---|---|
SRR606249 (podar) | 47.8% | 95.9% |
SRR1976948 (hu-s1) | 21.4% | 68.0% |
SRR12324253 (zymo mock) | 16.2% | 96.6% |
SRR5650070 (p8808mo11/iHMP) | 34.8% | 85.6% |
a few observations -
Note this is with genome-grist v0.9.0, so adapter and quality trimmed only.
a lot of our ancillary documentation and workflows (genome-grist and spacegraphcats) suggests doing k-mer trimming of metagenomes, e.g. with trim-low-abund.
but, for gather in particular, this is not particularly important. we should update documentation with a few points -
search
andsearch --containment
andprefetch
) will be decreased by the presence of erroneous k-mers. If this is a problem for you, you should trim.sourmash gather
uses shared k-mers, and is robust to erroneous k-mers - in particular,p_match
to genomes will be robust b/c the denominator uses a genome size, and the numerator is shared k-mers and hence will ignore errors. ref https://github.com/sourmash-bio/sourmash/issues/1289sourmash gather
with an abundance-weighted query will perform better because low-abundance k-mers will have less impact on the final coverage estimate. ref https://github.com/sourmash-bio/sourmash/issues/1818@bluegenes this sort of fits into some of the things we've been realizing with respect to jaccard vs containment - containment is much more robust with respect to erroneous k-mers, in various simple but important ways.
ref trimming paragraph in https://github.com/sourmash-bio/sourmash/issues/1135: