Table 1.

Differential expression calls

EE	DEC	raw	sva	sva+FC	sva+FC+AE
r-Make	limma	7226	8078	4498 [56%]	3058 [38%]
	edgeR	7314	8720	4908 [56%]	3058 [35%]
	DESeq2	6974	8380	4552 [54%]	3060 [37%]
Subread	limma	9772	9557	4795 [50%]	3016 [32%]
	edgeR	10202	10522	5398 [51%]	3036 [29%]
	DESeq2	9308	9709	4662 [48%]	3052 [31%]
TopHat2/	limma	8854	8782	4450 [51%]	3058 [35%]
Cufflinks2	edgeR	7329	7104	4386 [62%]	3018 [42%]
	DESeq2	8536	8489	4077 [48%]	3061 [36%]
SHRiMP2/	limma	8952	8276	4086 [49%]	3045 [37%]
BitSeq	edgeR	8791	8663	4526 [52%]	3025 [35%]
	DESeq2	7590	7878	3804 [48%]	3038 [39%]
kallisto	limma	8984	8851	4410 [50%]	3022 [34%]
	edgeR	9356	9284	4666 [50%]	3039 [33%]
	DESeq2	8016	8296	3915 [47%]	3044 [37%]

The table displays the number of differential expression calls, reflecting sensitivity, as obtained after specific analysis steps. For all combinations of methods for expression estimation and differential expression calling, we compare the typical numbers of genes classified as differentially expressed (q<5%). The columns show median results across sites for: raw expression estimates; expression estimates after svaseq correction; expression estimates after svaseq correction and application of additional filters for effect strength, i.e., fold-change (|log2F C|>1); and expression estimates after svaseq correction and application of additional filters for effect strength (|log2F C|>1) and minimum average expression (AE thresholds in Table 2). The last two columns also give a percentage relative to the numbers of genes found after svaseq correction and no additional filters. This highlights that the additional filtering for weak expression removes only a further 16% of genes originally classified as differentially expressed in addition to the ones already removed by the usual filters for log-fold change, affecting just 2.5% of all genes