Hi, I see the advantage of using lfcshrink in DESeq2 analysis for getting a more "realistic" FC that takes into account the variance of the data.
I have a question regarding what variable generated by DESeq2 to use for the post-DESeq2 analysis.
I run GSEA based on DESeq2 results in order to identify signatures enriched in one condition.
The input data to GSEA is a table of genes associated to a value that are ordered from the gene the most upregulated to the more downregulated in one condition compared to another, according to this "value".
Till now, I used the "stat" column generated by DESeq2 using the normal method for the comparison of two conditions.
However if i prefer the lfcshrink FC (apeglm or ashr methods) then what would be the value to be used for GSEA? Indeed the "stat" column is generated using the normal method and not lfcshrink methods, hence genes are not ordered similarly depending on the method used, and using the normal method to do GSEA while presenting the lfcshrink FC for individual gene comparison does not seem coherent for me.
Thanks for your response, Best regards Delphine
Both are fine, though for many that I talked to the shrunken logFCs are a more intuitive choice and probably a more direct measurement of the effect size than the pvalue (which is basically the stat column). Unfortunately, neither measures interestingness. Our life could be so easy if they did.