I run RNA-seq differential expression with DESeq2, followed by functional enrichment using clusterProfiler's gseGO and gseKEGG.

My previous setup:

Input: only significant DEGs, ranked by log2 fold change pAdjustMethod = "none", pvalueCutoff = 0.1

Recent reading made me realize that this may be incorrect.

My proposed setup:

Input: the full ranked gene list pAdjustMethod = "BH", pvalueCutoff = 0.05

My questions:

Should GSEA use the full ranked gene list rather than a filtered DEG subset? My understanding is that using only DEGs violates the method's assumptions. Moving from no correction to BH substantially reduces the number of significant terms. Is this expected because the previous results were inflated by false positives? After BH correction, KEGG still returns usable results, while GO BP often drops sharply or returns zero terms. Is this mainly because GO tests many more redundant terms? Is simplify() or a relaxed threshold such as q < 0.1 commonly used? For weak contrasts, is it appropriate to report that no pathways passed FDR q < 0.05, while showing top nominal terms only as suggestive and not significant?

Thanks. I mainly want to confirm that the revised approach is standard before using it for further analysis.