Cross-platform Bayesian deconvolution of MeDIP-seq reveals tissue-specific and cancer-associated methylation signatures

Abstract DNA methylation profiling of cell-free DNA is increasingly used for tissue-of-origin analysis and cancer detection, but quantitative interpretation of enrichment-based sequencing data remains a major challenge. MeDIP-seq provides scalable, cost-effective profiling of low-input samples like cell-free DNA, but lacks the absolute methylation quantification required for cell type deconvolution. Here we show that a Bayesian hierarchical model integrating MeDIP-seq with reference methylation atlases derived from direct methylation profiling enables accurate cross-platform cell type deconvolution. We validate the decemedip model through simulations and matched cross-platform datasets and demonstrate its ability to identify tissue-specific and cancer-associated methylation signatures in patient-derived xenografts and cell-free DNA. Our findings establish a quantitative framework for interpreting enrichment-based methylation sequencing data, with potential translational impact in noninvasive cell-free DNA-based diagnostics. decemedip is available at https://bioconductor.org/packages/decemedip/ .