Table 2.

Gene-level accuracy comparison. The table gives Pearson correlations between true log₂-expression levels and log₂-FPKM values produced by each workflow. The align-F + featureCounts workflow gives the best correlation in each case

Workflow	UHRR	HBRR	Simulation
align-F + featureCounts	0.851	0.870	0.955
align-G + featureCounts	0.850	0.869	0.955
STAR + featureCounts	0.848	0.867	0.901
STAR + htseq-count	0.845	0.864	0.877
STAR + summarizeOverlaps	0.845	0.864	0.877
TopHat2 + htseq-count	0.843	0.863	0.921
TopHat2 + summarizeOverlaps	0.843	0.863	0.921

Columns ‘UHRR’ and ‘HBRR’ are for the SEQC UHRR and SEQC HBRR samples respectively. For the SEQC columns, the log₂-expression values of 958 genes measured by TaqMan RT-PCR are taken as true values. Column ‘Simulation’ shows simulation results for 28 395 genes. For all columns, an offset of 1 was added to raw gene counts to avoid taking logarithms of zeros.