New to DESeq2.

I am using DESeq2 to normalize data for some microbiome and eDNA metabarcoding projects. My datasets are observational/ecological and there are many factors that I am interested in assessing with downstream analyses. For example, with my microbiome dataset, I am interested in understanding differences between sexes, age groups, reproductive status, and season of certain wildlife species. However, I'd also like to do more analyses with machine learning on other ecological factors.

When running DESeq2, should I incorporate all factors of interest in the design (e.g., ~ sex + age + reproductive status + season)? Or is there a way to run the design on a different factor (like sequencing run) so that I can then use the data that has been normalized across runs to do my downstream analyses on the other factors?

Thanks

I'll answer the design question first, and then make a note about DESeq2 for microbiome data:

1) It's good to always include the covariates that may explain variance in counts (if numeric, first center and scale them, factors are fine as is) when testing particular covariates, with the exception of trying to make causal claims, and then you need to consider the DAG. But if you are just testing for associations, just always include the covariates you believe are explanatory. If there are too many you can reduce their dimension with various techniques. But with four you should be fine to just include them, as long as the experimental design is well balanced.

2) I've learned from microbiome statistical folks that they really don't like to use DESeq2. I don't work in microbiome, but I suppose there are other methods that would be a better match for the distributions seen in micriobiome / metagenomics. As heard in Bioc2022, Amy Willis's lab at UW has done a good job with methods that deal better with e.g. normalization and relaxing the distributional assumptions: