miRglmm: a generalized linear mixed model of isomiR-level counts improves estimation of miRNA-level differential expression and uncovers variable differential expression between isomiRs

Table 2 Comparing performance of miRglmm to aggregation methods

	MSE (10⁻³)	Coverage proportion	Null variance (10⁻³)	abs(FC) = 2 variance^a (10⁻³)	TPR	TNR	AUC
miRglmm	7.81	0.98	5.01	6.92	0.95	0.97	0.99
miRglmm-Poisson	32.12	0.93	12.95	38.16	0.92	0.94	0.99
NB GLM	12.11	0.97	5.66	9.27	0.96	1.00	1.00
DESeq2	10.19	0.99	5.58	8.00	0.92	1.00	1.00
edgeR	12.11	NA	5.66	9.28	0.90	1.00	1.00
limma-voom	13.78	0.99	6.18	10.66	0.92	1.00	1.00

MSE mean squared error, DE differentially expressed, TPR true positive rate, TNR true negative rate, AUC area under ROC curve, NA not applicable, TPR by abs(FC) is provided in Supplemental Table 27
^avariance for all FC levels shown in Additional File 1: Table S28

ISSN: 1474-760X