DRPreter: Interpretable Anticancer Drug Response Prediction Using Knowledge-Guided Graph Neural Networks and...

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Drug sensitivity prediction models often use graph neural networks (GNNs) for drug structure or gene networks, or focus on model interpretability.
• Integrating knowledge-guided approaches with interpretability is crucial for enhancing prediction accuracy and practical application of drug response models.

Purpose of the Study:

• To develop an interpretable and knowledge-guided model, DRPreter (drug response predictor and interpreter), for predicting anticancer drug response.
• To improve drug response prediction accuracy and enhance the practical utility of predictive models through enhanced interpretability.

Main Methods:

• DRPreter utilizes graph neural networks (GNNs) to learn cell line and drug information.
• Cell-line graphs are decomposed into subgraphs based on biological pathway domain knowledge.
• A type-aware transformer identifies pathway-drug relationships, highlighting key pathways in drug response.

Main Results:

• DRPreter significantly outperforms existing state-of-the-art graph-based models on the Genomics of Drug Sensitivity and Cancer (GDSC) dataset.
• The model successfully identified key genes and pathways for specific drug-cell line pairs, with findings supported by existing literature.
• Experimental results demonstrate superior performance in drug response prediction compared to current methods.

Conclusions:

• DRPreter offers a novel approach to knowledge-guided and interpretable anticancer drug response prediction.
• The model's ability to identify key pathways and genes provides insights into drug action mechanisms.
• DRPreter enhances the practical application of predictive models by offering interpretable predictions for drug sensitivity.