A fuzzy method for RNA-Seq differential expression analysis in presence of multireads

BMC Bioinformatics

Table 6 Comparisons of the results obtained on simulated data

Induced variation of expression	Gene	Cuffdiff + TopHat (p-value <0.05)	DESeq2 + Centroids (p-value <0.05)	DESeq2 + RSEM (p-value <0.05)	edgeR + Centroids (p-value <0.05)	edgeR + RSEM (p-value <0.05)	Same-expression Possibility	Log2 Fold Change on centroids
Under	FADS2			*	**	**	0.000	−1.01
Under	PDE4DIP			*	**	**	0.529	−0.98
Under	U2AF1			*	**	**	0.989	−0.82
Under	WAC			*	**	**	0.000	−1.00
Under	CLCA1			*	**		0.000	−0.97
Under	S100A8				**	**	0.000	−1.04
Under	BPG254F23.5			*			0.889	−0.88
Under	EEF1A1			*			0.889	−0.95
Under	LYZ			*			0.778	−0.91
Under	SCGB1A1						0.000	−0.97
Over	HLA-DQB1			*	**	**	0.778	0.99
Over	IFI6			*	**	**	0.000	0.97
Over	IGF1R			*	**	**	0.000	1.00
Over	MUC5AC			*	**	**	0.000	1.01
Over	POSTN			*	**	**	0.000	1.00
Over	CD177			*	**		0.222	0.99
Over	HLA-DRA			*	**		0.794	0.94
Over	NSUN5P1			*		**	0.889	0.90
Over	UBBP4			*			0.778	0.98
Over	MTRNR2L12						0.778	0.96
None	DUX4L7			*			1	−0.02
None	CH507-152C13.1			*	*		0.995	−0.60
None	AC016698.1			*			1	−0.06
None	MRPL51P2					**	1	−0.05
None	CH507-152C13.4					**	1	0.05
TP	True Positives	0	0	17	13	11
TN	True Negatives	20,000	20,000	19,997	19,999	19,998
FP	False Positives	0	0	3	1	2
FN	False Negatives	20	20	3	7	9
TP/(TP + FN)	Sensitivity (recall)	0 %	0 %	85 %	65 %	55 %
TN/(TN + FP)	Specificity	100 %	100 %	99.9 %	100 %	99.9 %
TP/(TP + FP)	Precision	0 %	0 %	85 %	93 %	85 %

This table lists the results obtained on 10 over-expressed and 10 under-expressed genes. Five more genes are involved as false positives. Some workflows select the genes as significant DE events only by the p-value (*), while others compute also the correct FC (**). The first two workflows are not able to select significant p-values. Ten genes have a low uncertainty (same-expression possibility) and ten genes have a high uncertainty due to their multireads. The five false positive events have a same-expression possibility close to 1 and a non significant FC. Even in this case, there is an overall disagreement in the selection of DE genes. The last 7 rows of the table show some evaluation metrics about the results (Bold)

ISSN: 1471-2105