DualGCN: a dual graph convolutional network model to predict cancer drug response

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Drug resistance poses a significant challenge in cancer therapy, necessitating improved methods for predicting anti-cancer drug responses.
• Cancer cell line data offers valuable insights but doesn't fully represent tumor microenvironments, highlighting the need to bridge in vitro findings with clinical and single-cell data.
• Current predictive models often rely on single nucleotide variants (SNVs), which are difficult to obtain reliably from clinical and single-cell samples, limiting their applicability.

Purpose of the Study:

• To develop a novel computational method for predicting cancer drug response that overcomes limitations of existing SNV-based approaches.
• To create a unified model that integrates drug chemical structures and biological sample omics data for enhanced predictive accuracy.
• To enable the generalization of drug response predictions from cell line studies to clinical and single-cell tumor data.

Main Methods:

• Introduced DualGCN, a Dual Graph Convolutional Network model for predicting cancer drug response.
• Encoded drug chemical structures and biological sample omics data using graph convolutional networks.
• Integrated prior knowledge of cancer-related genes and protein-protein interactions into the model.

Main Results:

• DualGCN effectively predicts cancer drug response by encoding both drug chemical structures and omics data.
• The model demonstrated superior performance compared to most state-of-the-art methods.
• DualGCN successfully avoided the need for large-scale single nucleotide variant (SNV) data, enhancing its applicability to diverse datasets.

Conclusions:

• The developed DualGCN method shows significant promise for predicting cancer drug response, outperforming existing approaches.
• The model's ability to function without SNV data makes it suitable for extension to clinical and single-cell tumor samples.
• These findings represent a potential advancement in precision medicine by improving anti-cancer drug treatment strategies.