Fast and robust adjustment of cell mixtures in epigenome-wide association studies with SmartSVA

BMC Genomics

Table 1 EWAS summary statistics for two real data sets. FDR control and Bonferroni correction were used for selecting DMPs

Data set	Method	π ₀ ^a	λ ^b	#(q < 0.05)^c	#(p_b < 0.05)^d
GSE30601 (Gastric cancer, 27 K)	Unadjusted	0.18	25.8	21,487	8,323
	SVA^e	0.13	59.3	23,404	13,846
	SmartSVA	0.75	1.72	888	30
	RefFreeEWAS	0.70	2.03	1,266	68
	EWASher^f	1.00	0.87	3	1
	ReFACTor	0.95	1.07	23	1
GSE40279 (Human aging, 450 K)	Unadjusted	0.35	7.73	245,279	41,357
	SVA	0.60	2.98	102,509	20,644
	SmartSVA	0.87	1.33	5,620	679
	RefFreeEWAS	0.68	2.20	43,791	5,192
	ReFACTor	0.91	1.23	1,577	177

^a π ₀ is the percentage of non-DMP estimated based on “qvalue” method
^b λ is the genomic inflation factor calculated on all CpGs
^c FDR control is based on “qvalue” method and 5% level
^dBonferroni correction was used at 5% level
^eThe classic SVA with default implementation was used (B = 5)
^fFaST-LMM-EWASher was performed without filtering out consistently methylated or unmethylated CpGs. For the age data set, we were unable to obtain the results within one week

ISSN: 1471-2164