Ensemble-based method for detecting differential expression in RNA datasets provides greater sensitivity and less biased p-values than existing methods

In examining two-group differential analysis, it is apparent that different statistical methods can yield very different results even given the same RNA-Seq input data. In particular, we have observed that parametric methods such as t-test and limma yield very similar results which are very different than methods that are based on count based testing such as edgeR and DESeq2 (with edgeR and DESeq2 also yielding very similar results). Still another approach based on empirical Bayes modeling that seemingly provides yet another distinct set of information is EBSeq . In our approach, we implement “stacking”, an ensemble-based method using a meta-level model to adjudicate a subset of the traditional group testing methods and combine information from different statistical testing approaches. This ensemble method provides less biased p-values and yields superior AUC performance when determining differential gene expression (DGE).


Complete the form to access this whitepaper.