The "inverse pvalue" computed with
1 - padj is problematic with R due to rounding problem when pvalue is next to 0.
> 1 - 1.32e-30  1 > 1 - 1.32e-30 == 1  TRUE
This can cause problem when ranking gene list for gsea analysis, because many inverse pvalues become equal.
Computing it with
1 / padj (the true inverse function) solve this issue.