Table 1.

Effect of various bias correction methods on methylation concordance for 19 pairs of duplicate samples evaluated using mean-centered correlation coefficient (larger is better) and average absolute methylation difference (smaller is better)

	Infinium I				Infinium II
Method	Ra	Pb	MethDiffc	Pb	Ra	Pb	Methdiffc	Pb
RAW	0.449	Ref	2.182	Ref	0.387	ref	3.165	ref
BMIQ	0.449	Ref	2.182	Ref	0.428	0.024	2.782	$9.1\times {10}^{-3}$
SWAN	0.446	0.625	2.223	0.900	0.436	$7.3\times {10}^{-3}$	2.499	$3.5\times {10}^{-4}$
RCP	0.449	Ref	2.182	Ref	0.500§	$6.5\times {10}^{-4}$	2.354‖	$2.8\times {10}^{-4}$

Although all methods provide improvement over unadjusted type II probe raw data, RCP is significantly better than BMIQ and SWAN.

^aAverage mean-centered correlation coefficient for 19 duplicate-pairs.

^bBased on one-sided Student’s paired T-test against raw result.

^cMethDiff: Average mean absolute methylation beta value difference (%) for 19 duplicate pairs.

§P-values of one-sided Student’s paired T-test of RCP result against BMIQ result and SWAN result are $4.8\times {10}^{-3}$ and $2.9\times {10}^{-4}$ respectively.

^‖P-values of one-sided Student’s paired T-test of RCP result against BMIQ result and SWAN result are $2.8\times {10}^{-4}$ and 0.019 respectively.