SCMG: A Multi-Omics and Stage-Resolved Single-Cell Framework for Cancer Driver Gene Prioritization

Abstract Background Cancer driver gene prioritization largely relies on bulk molecular profiles, leaving epithelial transcriptional changes across disease stages underused. Results We present Single-Cell Multi-omics Graph (SCMG), a graph-learning framework integrating bulk mutation, DNA methylation and gene expression with stage-resolved epithelial single-cell data and biological interaction networks. SCMG outperformed five comparison methods across six networks and achieved the highest area under the precision–recall curve in nine of 15 cancer types and the highest area under the receiver operating characteristic curve in seven. Ablation analyses supported contributions from single-cell features, network propagation and multimodal integration. External catalogue evaluation showed enrichment for driver genes outside the training positive-label set. Combining prediction scores with a background-corrected measure of stage-associated expression change organized high-scoring candidates into two groups with contrasting expression patterns. Their gene-expression signatures were associated with patient outcomes, while an external cutaneous squamous cell carcinoma cohort supported disease-associated expression of selected candidates. Conclusions SCMG provides a framework for integrating stage-resolved epithelial context into cancer driver gene prioritization and interpreting candidate rankings.