MAGIC: Methylation Analysis with Genomic Inferred Contexts

MotivationWhole-genome bisulfite sequencing (WGBS) enables base-resolution methylation profiling but poses statistical challenges for differential methylation analysis. While existing methods have made important progress using uniform dispersion shrinkage across CpGs, there remains opportunity to better capture methylation heterogeneity across the genome. There is a need to account for context-specific methylation variability and to improve detection accuracy under realistic sequencing constraints. ResultsWe propose MAGIC (Methylation Analysis with Genomic Inferred Contexts), a beta-binomial mixture model that learns genomic contexts directly from methylation data. MAGIC models genomic heterogeneity through context-specific dispersion shrinkage and detects differential methylation via two complementary tests, including a Wald test and Bayes factor test. Simulations across various coverage and effect sizes demonstrate that MAGIC improves sensitivity and false-positive rate compared with DSS, providing a robust framework for differential methylation analysis in WGBS data. AvailabilityThe source code of MAGIC is available at https://github.com/Pickledzebra/MAGIC. ContactMatteoP@mcdb.ucla.edu Supplementary informationSupplementary data are available at bioRxiv.