ScholarWorks@성균관대학교

초록

Cancer drug response prediction plays a crucial role in advancing precision medicine and anticancer drug development. While multi-omics integration has become a key trend, existing methods struggle to effectively model the complex nonlinear relationships between cancer cell line multi-omics data and drug chemical properties. Furthermore, current models lack fine-grained cross-modal interactions, limiting their interpretability and clinical translatability. To address these limitations, we propose DeepKGI, a drug response prediction framework based on cross-layer graph fusion and key gene identification. DeepKGI integrates drug molecular structural features through a cross-layer graph feature fusion mechanism and employs an attention-driven cross-modal fusion strategy to model complex interactions within multi-omics data. Compared to existing approaches, this method enables more comprehensive extraction of drug-related features and improved integration of multi-omics data. Additionally, DeepKGI includes a key gene identification component that uncovers gene-level regulatory associations contributing to drug response, enhancing the biological interpretability of the model. Experiments on benchmark datasets demonstrate that DeepKGI achieves a Pearson correlation of 0.945 and an RMSE of 0.899, outperforming state-of-the-art methods across multiple evaluation metrics. Case studies further confirm its capability to accurately predict drug sensitivity and identify potential key target genes, providing actionable biological insights for personalized cancer therapy and novel drug discovery.

키워드