From multi-omics data to the cancer druggable gene discovery: a novel machine learning-based approach.

This article focuses on DF-CAGE, a novel machine learning method designed to identify cancer-druggable genes by integrating multi-omics data, including somatic mutations, copy number variants (CNVs), DNA methylation, and RNA sequencing from approximately 10,000 profiles in The Cancer Genome Atlas (TCGA). DF-CAGE employs a multi-granularity scanning module and a cascade forest module to capture complex nonlinear relationships across these data types, achieving high predictive performance (AUROC ~0.9) on benchmark datasets such as OncoKB, TARGET, and DrugBank. The method identified 465 potential cancer-druggable genes (CDG), categorized into known, reliable, and candidate sets, with many overlapping established cancer driver genes but also revealing novel targets beyond current knowledge. Feature importance analysis indicated that CNV-related data, gene network connectivity, and mutation rates are primary contributors to druggable gene identification, highlighting the value of multi-omics integration for advancing precision oncology and drug development.