DualGCN

DualGCN predicts cancer drug response by integrating drug chemical structures and gene features from cancer samples to estimate half-maximal inhibitory concentration (IC50).

Key Features:

• Graph Convolutional Networks (GCNs): Uses GCNs to encode chemical structures of drugs and omics-derived features from biological samples.
• Integration of Omics Data: Incorporates gene features and prior knowledge on cancer-related genes and protein-protein interactions to inform representations.
• Multilayer Perceptron (MLP): Combines embeddings from drug and sample encoders and processes them through an MLP to predict response.
• IC50 Output: Produces half-maximal inhibitory concentration (IC50) values as the quantitative measure of drug response.
• Avoidance of Large-Scale SNV Data: Operates without requiring extensive single nucleotide variant (SNV) datasets, enabling applicability to clinical tumor and single-cell samples.

Scientific Applications:

• Improving Anti-Cancer Drug Treatment: Predicts drug sensitivity to support selection and optimization of therapeutic regimens.
• Guiding Drug Design: Provides molecular response insights that can inform development of compounds with improved efficacy or reduced resistance.
• Precision Medicine Advancements: Generalizes beyond cell line data to clinical tumor and single-cell contexts to support personalized oncology approaches.

Methodology:

Employs GCNs to encode drug chemical structures and omics-derived gene features (including cancer-related genes and protein-protein interactions), combines resulting embeddings, and applies an MLP to predict IC50 while avoiding reliance on large-scale SNV data.