pval adjustment

[leda]

Why do the DESeq2/DEXSeq vignettes use p-value adjustment by the "BH" method, instead of the "BY" method? Shouldn't the latter be better for possibly-dependent gene counts, or does that turn out to be overly conservative in practice?

thanks

[Simon Anders]

The latter.

BH is fine even in case of dependence between the tested hypothesis as long as the dependence structure is has a certain property called "positive regression dependence". This is a quite technical definition but you are usually safe as long as all correlations between genes are expected to be zero or positive, which is a reasonable assumption to make.