MCMSeq: Bayesian hierarchical modeling of clustered and repeated measures RNA sequencing experiments

BMC Bioinformatics

Table 5 Number of significant genes by contrast in the tuberculosis dataset. 15,993 genes were included in the analysis. Genes were called significant if the FDR adjusted p-value was <0.05

	edgeR	edgeR ^∗	limma	LMM	MCMSeq	NBGLMM	ShrinkBayes¹
Number Failed to Converge	0	0	0	392	132	1348	0
Baseline Differences
Latent vs. Control	279	-	11	2	3	9	416, 407
Progressor vs. Control	2221	-	1009	752	650	1130	1821, 1013
Progressor vs. Latent	1643	-	552	299	256	667	1369, 708
Changes Over Time
Control	1	0	0	0	0	0	2977
Latent	2	0	0	0	3	2	3171
Progressor	664	790	2460	1413	1463	1878	7848
Differences in Changes
Over Time
Latent vs. Control	0	0	0	0	3	0	3331, 3230
Progressor vs. Control	16	272	310	537	635	794	7199, 6429
Progressor vs. Latent	170	278	1852	825	852	1328	6681, 7390

¹In order to test all contrasts of interest, ShrinkBayes models were fit three times, using a different reference group each time. This resulted in some contrasts being evaluated twice. As ShrinkBayes is a stochastic method, the number of significant genes can differ between model fits. We present the results from both models when tests were performed multiple times

ISSN: 1471-2105