Journal of Biometrics & Biostatistics

The role of genome-wide patterns of methylation in human disease has drawn attention increasingly in recent  years, because the methylome has the potential for large effects in disease etiology. Most analyses of methylation have utilized the percent signal that is methylated, known as β-value, or the logistic transformation of β, named M-value, as the summary measures. However, in general, these summary measures do not follow a Normal distribution and lead to statistical tests sensitive to outlying samples. In this paper, we proposed the N-value, a type of weighted logistic transformation of β that accounts for signal variability among beads for analyses of differential methylation. Our analysis of 27K Illumina array data showed that the N-value follows a desirable shape of sample distribution, and its test is robust to outliers. Through a simulation study, we presented results that show the t-tests of the N-value is more consistent, and has greater power under the presence of heterogeneity of samples and in different sample sizes.