Precision Another important consideration from the perspective of an investigator is the precision, or fraction of true positives in the set of genes which pass the adjusted P value threshold. This can also be reported as 1−FDR. Again, ‘true’ differential expression was defined by an adjusted P value <0.1 in the larger verification set. The estimates of precision are displayed in Figure 9, where we can see that DESeq2 often had the second highest median precision, behind DESeq (old). We can also see that algorithms with higher median sensitivity, e.g., DSS, were generally associated here with lower median precision. The rankings differed significantly when Cuffdiff 2 was used to determine the verification set calls. This is likely due to the additional steps Cuffdiff 2 performed to deconvolve changes in isoform-level abundance from gene-level abundance, which apparently came at the cost of lower precision when compared against its own verification set calls.
