Entering edit mode
anjali.kumari
▴
10
@anjalikumari-21851
Last seen 4.6 years ago
Hi All,
I would like to do differential abundance analysis using DESeq2 for shotgun metagenome data. I have used the MOCAT2 for generating functional annotations. Eg: For ARDB database, Do we need to consider the "unassigned reads" (reads which didn't get assigned to any of the ARDB categories - which is around 99% of the reads) for DESeq2 analysis? Will "unassigned reads" impact the sizefactorestimation calculation?
This is an active question in the community. Many papers use DESeq2. I am doing microbiome pathway analysis, and my KO and pathway abundances are mostly slightly bimodal, and certainly do not follow a negative binomial distribution (what the DESeq2 model is built around).
In discussions, I've had a few people suggest simply relying on non-parametric group tests. That does not work for us because we have a large number of covariates--we need to use something that works like regression or ancova. But those are dangerous with the kind of distributions I'm seeing--hence DESeq2 seems tempting because it works very hard to do proper normalization.
I have been considering PERMANOVA, which I have used for beta-diversity. But I do not know if it is appropriate for this job.
Cross-posted https://www.biostars.org/p/397969/