Analysis of DNA methylation epidemiological data through a generic composite statistical framework

DNA methylation events represent epigenetic heritable modifications that regulate gene expression by affecting chromatin remodeling. They are encountered more often in CpG rich promoter regions, while they do not alter the DNA sequence itself. High-volume DNA methylation profiling methods exploit microarray technologies and provide a wealth of data. This data solicits rigorous, generic, yet ad-hoc adjusted, analytical pipelines for the meaningful systems-level analysis and interpretation. In this work, the Illumina Infinium HumanMethylation450 BeadChip platform is utilized in an epidemiological cohort from Italy in an effort to correlate interesting methylation patterns with breast cancer predisposition. The composite computational framework proposed here builds upon well established, analytical techniques, employed in mRNA analysis. For analysis purposes, the log2(ratio) of the intensities of a Methylated probe (I_Meth) versus an UnMethylated probe (I_Un-Meth), quoted as M-value, is used. Intensity based correction of the M-signal distribution is systematically applied, based upon Intensity-related error measures from quality controls samples incorporated in each chip. Thus, batch effects are corrected, while probe-specific, intensity-related, error measures are considered too. Robust, (based on bootstrapping) statistical measures measuring biological variation at the probe level, are derived in order to propose candidate biomarkers. To this end, coefficient variation measurements of DNA methylation between controls and cases are utilized, alleviating simultaneously the impact of technical variation, and are juxtaposed to classical statistical differential analysis measures.