A comparison of reference-based algorithms for correcting cell-type heterogeneity in Epigenome-Wide Association Studies

Table 2 Comparison of the relative sensitivity and specificity. Smoking-associated DMCs (sDMCs) were defined at FDR < 0.05 and FDR < 0.3 and for each of 5 methods in the 152 whole blood Illumina 450k samples from the MRC1946 birth cohort data

nTP = 62, nTN = 89290	NoADJ	EpiD (noDHS)	EpiD	CBS	CP
FDR < 0.05
sDMCs	34	67	70	58	152
\| TP ∩ sDMC \|	31	34	35	36	37
Sensitivity	0.50	0.55	0.56	0.58	0.60
FP = \| TN ∩ sDMC \|	0	3	3	0	11
Specificity	1	0.999966	0.999966	1	0.999877
Empirical FDR	<0.03	0.045	0.043	<0.02	0.072
FDR < 0.3
sDMCs	67	544	802	259	10457
\| TP ∩ sDMC \|	40	50	52	44	54
Sensitivity	0.64	0.81	0.84	0.71	0.87
FP = \| TN ∩ sDMC \|	1	73	121	31	1735
Specificity	0.99998	0.9992	0.998	0.9996	0.98
Empirical FDR	0.01	0.13	0.15	0.12	0.17

Methods are: NoADJ = no-adjustment, EpiD (noDHS) = adjustment with RPC using reference with no DHS info, EpiD = adjustment with RPC and a reference with DHS info, CBS = adjustment with CIBERSORT, CP = adjustment with constrained projection. Sensitivity was estimated as the fraction of the 62 gold-standard true positives (TP) found among the sDMCs. In the case of specificity, this was estimated as the fraction of true negatives (TN) among sDMCs measured relative to a gold-standard set of 89290 true negative (TN) CpGs

ISSN: 1471-2105