Big differences in enricher results between clusterProfiler 4.16.0 and clusterProfiler 4.20.0

[ckeeling]

Hello @GuangchuangYu

I recently updated to R 4.6.0 and thus clusterProfiler 4.20.0 from 4.16.0. I am seeing a large decrease in the number of significant terms returned across several comparisons. Is this expected based on the bugs fixed and switching internally from DOSE::enricher_internal to enrichit compared to 4.16.0?

    pvalueCutoff = 0.05,
    pAdjustMethod = "BH",
    qvalueCutoff = 0.05,
    minGSSize = 1,
    maxGSSize = 1000

Thanks

[ckeeling]

Hello @GuangchuangYu , following up on my above comment, I examined multiple enricher results when using the two versions of clusterProfiler.

clusterProfiler * 4.16.0 2025-04-15 [1] Bioconductor 3.21 (R 4.5.0)
clusterProfiler * 4.20.0 2026-04-28 [1] Bioconductor 3.23 (R 4.6.0)

I then tried the devel version on Bioconductor:
clusterProfiler * 4.21.0 2026-05-18 [1] Bioconductor 3.24 (R 4.6.0)
Results were identical to the 4.20.0 version.

The differences can be quite large between 4.16.0 and 4.20.0/4.21.0. In addition, 4.16.0 found 12 comparisons producing significant terms that 4.20.0/4.21.0 did not identify any terms. Any ideas why they might differ so much, and which result should be considered correct?

Thanks,
Chris