Bonferroni and FDR multiple testing corrections too strict for differential expression analyses

Issue description

Calculations for the Bonferroni and FDR multiple testing corrections in Differential Expression for RNA-Seq and Differential Expression in Two Groups used an inflated value for the number of tests performed. In affected software versions, this number included both the number of tests performed and the number of untestable genes/transcripts, i.e. those with NaN as an expression value.

The impact of this is that fewer differentially expressed transcripts/genes are reported after applying a p-value cut-off based on these corrections than should have been the case. The missing transcripts/genes will be those nearest to the p-value cut-off, i.e. they will not be the most significantly differentially expressed transcripts/genes.

Even when using affected software:

• Transcripts/genes that are reported as differentially expressed are correctly reported.
• For sets of samples with no NaN values, the multiple testing corrections are correct.
• ‘Raw’ p-values are correctly reported.

Recommendations

For results generated using affected software versions, upgrading to an unaffected version is generally recommended. If analyses are then re-run using the tools Differential Expression for RNA-Seq and Differential Expression in Two Groups, an increased number of differentially expressed transcripts/genes may be reported due to the changes made to the multiple testing correction methods.

Affected versions

• CLC Genomics Workbench 10.0 to 12.0.2
• CLC Genomics Server 9.0 to 11.0.2
• Biomedical Genomics Workbench 5.0 and above
• The Advanced RNA-Seq plugin available for CLC Genomics Workbench 9.x and CLC Genomics Server 8.x

A fix was implemented in CLC Genomics Workbench 12.0.3 and CLC Genomics Server 11.0.2.