Identifying cancer driver genes is fundamental for understanding tumor initiation and guiding therapeutic strategies. However, most existing methods assess gene importance from a global or static perspective, overlooking sample-specific functional differences in the same gene. To address this, we propose TGBWDriver, which integrates a two-layer GraphSAGE with bidirectional weighted feature aggregation to capture structural characteristics while distinguishing context-dependent gene functions. An exponential pairwise voting strategy prioritizes candidate driver genes, improving ranking stability and accuracy. Systematic experiments on BRCA, LUAD, and PRAD datasets show that TGBWDriver outperforms five existing methods in precision, recall, and F1-score. Ablation studies confirm the critical role of each component. Moreover, TGBWDriver demonstrates strong capability in identifying potential novel cancer driver genes, with predictions showing significant biological relevance in GO enrichment and KEGG pathway analyses. The method provides an effective computational framework for cancer driver gene identification. The source code and datasets are freely available at https://github.com/SCSMDyeah/TGBW Accessed on 4 May 2026.
Chen et al. (Tue,) studied this question.