Comparison of different cell type correction methods for genome-scale epigenetics studies

BMC Bioinformatics

Table 2 Number of significant CpG sites with and without cell-correction methods and overlap of CpG sites with those from the SVA method (example data from FasT-LMM-EWASher package)

Method	Identified CpGs (N)^#	Overlap with SVA (%)	p-value^##	J-index ^c
Houseman et al.	1835	54.71	<0.0001	0.40
minfi	3589	84.59	<0.0001	0.40
SVA	1888	–	–	–
RefFreeEWAS	788	30.51	<0.0001	0.30
RefFreeCellMix	1006	18.38	<0.0001	0.10
ReFACTor	4224	87.45	<0.0001	0.40
EWASher ^a	3	0.16	<0.0001	0
RUV	6008	99.95	<0.0001	0.30
Unadjusted ^b	3768	82.89	<0.0001	0.40

^#The selection of CpG sites is based on FDR-adjusted p-values (FDR is controlled at 0.05)
^## P-value is based on Fishers exact test for overlap. The null hypothesis is that there is no overlap with the CpGs identified based on SVA
^aThe FasT-LMM-EWASher method
^bUnadjusted: cell type compositions were not incorporated into the analyses
^cJ-index is Jaccard index

ISSN: 1471-2105