MDAGCN: Predicting Mutation-Drug Associations Through Signed Graph Convolutional Networks via Graph Sampling.

Area of Science:

• Computational biology
• Genomics
• Pharmacogenomics

Background:

• High-throughput molecular data in cancer precision medicine poses computational challenges.
• Genetic mutations can serve as biomarkers for predicting targeted drug responses.
• Accurate prediction of mutation-drug associations is crucial for cancer therapy and drug discovery.

Purpose of the Study:

• To propose a novel graph convolutional network method, MDAGCN, for predicting mutation-drug associations (sensitivity/resistance) in cancer.
• To enhance the efficiency and accuracy of computational models for mutation-drug interaction prediction.

Main Methods:

• Constructing a feature and topological graph using the k-Nearest Neighbors algorithm.
• Incorporating structural relationships and feature data of mutation-drug interactions.
• Utilizing a graph convolutional network (MDAGCN) for prediction.
• Employing a graph sampling technique for training signed graphs.

Main Results:

• MDAGCN demonstrates superior performance compared to state-of-the-art methods in predicting mutation-drug associations.
• The effectiveness of the graph sampling technique for training signed graphs was validated.
• The model accurately predicts drug sensitivity and resistance.

Conclusions:

• MDAGCN provides a comprehensive end-to-end framework for cancer pharmacogenomics.
• The framework facilitates the discovery of novel mutation-drug associations.
• It aids in the in-depth analysis of drug sensitivity and resistance in cancer treatment.