Housekeeping gene normalization is never applied by the tool Differential Expression in Two Groups

Issue description

The issue described here affects the tool Differential Expression in Two Groups

In affected software, normalization always proceeds using Normalization method = TMM, even when Normalization method = Housekeeping genes has been selected as in the screenshot below.

Effect on results

The choice of normalization method affects the calculated fold changes and p-values. In many cases, the two normalization methods will give consistent results, such that there is a very large overlap in the number of differentially expressed genes (defined by a combination of fold change and p-value cut-off).

However, results can differ substantially when the assumptions of either normalization method are violated:

• For TMM normalization, it is assumed that most genes are not differentially expressed. This assumption is especially likely to be violated for targeted panel data where the panel contains few genes, or where a large proportion of genes are likely to be differentially expressed.
• For housekeeping gene normalization, it is assumed that the housekeeping genes have stable expression across samples.

Recommendations

It is recommended to re-run affected data with the tool Differential Expression for RNA-Seq, which is not affected by this issue. Old and new results can be compared using Create Venn Diagram for RNA-Seq.

Affected software

This issue affects:

• QIAGEN CLC Genomics Workbench 12.0 through to 21.0.4
• QIAGEN CLC Genomics Server 11.0 through to 21.0.4

This issue was addressed in version 21.0.5 of the software.