Prediction of Cancer Drug Response Based on Hypergraph Convolutional Network and Contrastive Learning.

Area of Science:

• Computational biology
• Genomics
• Drug discovery

Background:

• Accurate prediction of cancer drug responses is crucial for personalized cancer treatment.
• Existing models struggle with generalization and robustness due to complex drug-cell line interactions.

Purpose of the Study:

• To develop a novel computational model, HypergraphCDR, for improved cancer drug response prediction.
• To enhance the generalization and robustness of drug response prediction by modeling higher-order drug-cell line relationships.

Main Methods:

• Proposed HypergraphCDR, a Hypergraph Convolutional Network with Hypergraph Contrastive Learning.
• Compressed multiomics features using an autoencoder and constructed a hypergraph to capture high-order relationships.
• Jointly optimized drug and cell line embeddings using supervised regression and contrastive loss.

Main Results:

• HypergraphCDR consistently outperformed state-of-the-art methods in PCC, SCC, and R².
• Demonstrated superior generalization performance on unseen drugs, cell lines, and in tissue-specific evaluations.
• The model showed enhanced accuracy and robustness in predicting cancer drug responses.

Conclusions:

• HypergraphCDR effectively models higher-order drug-cell line relationships, improving prediction accuracy and robustness.
• This framework supports reliable drug screening and treatment strategy development in precision medicine.
• The study offers a generalizable computational approach for advancing precision oncology.