I have a methodical question concerning the s-values computed by DESeq2::lfcShrink() using the apeglm shrinkage option: Why did you decide to calculate the average instead of the maximum of the local false sign rate (local FSR) of all genes which have a local FSR smaller than or equal to the local FSR of the gene under consideration? Likewise, if one sets an LFC threshold in DESeq2::lfcShrink(): Why did you decide to calculate the average instead of the maximum of the local false-sign-or-smaller (local FSOS rate) of all genes which have a local FSOS rate smaller than or equal to the local FSOS rate of the gene under consideration?

If you think about the expected number of binary events, that motivates the cumulative sum of the posterior probabilities. But also we were following precedent of some of the others working with a set of the smallest posterior probabilities. In the paper we have: "Other methods have suggested using the cumulative average or the cumulative maximum of posterior probabilities for defining the set of interesting features in high-throughput experiments include Leng et al. (2013), Choi et al. (2008), and Kall et al. (2008)."