A multi-layer framework for computational cancer epigenomics

Abstract

Cancer epigenomics has become central to understanding tumor initiation, progression, heterogeneity, and therapeutic response. High-throughput profiling technologies including bisulfite sequencing, chromatin immunoprecipitation sequencing (ChIP-seq), assay for transposase-accessible chromatin using sequencing (ATAC-seq), and RNA sequencing (RNA-seq) generate complex, multi-dimensional datasets that require robust computational frameworks for meaningful interpretation. This review outlines key bioinformatics workflows in cancer epigenomics, including data preprocessing, quality control, sequence alignment, signal detection, and differential analysis. While epigenomic data provide a mechanistic regulatory foundation, their full interpretive value emerges through integration with genomic, transcriptomic, and clinical data within computational oncology frameworks. Accordingly, we emphasize integrative modeling approaches that combine multi-omics data to uncover regulatory mechanisms, identify biomarkers, and define disease-associated molecular subtypes. Machine learning methods are increasingly applied for classification, prognosis prediction, and therapeutic response modeling; however, challenges remain in model interpretability, reproducibility, and external validation. We further highlight critical analytical limitations, including data heterogeneity, tumor complexity, lack of standardized workflows, and the persistent gap between association and biological mechanism. Emerging advances in single-cell epigenomics, spatial profiling, and explainable AI offer new opportunities to refine biological insight and clinical translation. Importantly, we propose a structured multi-layer interpretation framework that links computational outputs across data-level processing, epigenomics-informed integrative regulatory modeling, and multi-omics-informed clinical interpretation. This framework differs from existing pipelines by explicitly constraining how information is transformed across analytical layers, enabling traceable and mechanistically interpretable clinical inference.