HGCLMDA: Predicting mRNA-Drug Sensitivity Associations via Hypergraph Contrastive Learning

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Identifying mRNA-drug sensitivity associations is vital for drug development and disease treatment.
• Traditional experimental verification methods are time-consuming and labor-intensive.

Purpose of the Study:

• To develop HGCLMDA, a hypergraph contrastive learning approach for predicting mRNA-drug sensitivity associations.
• To improve the efficiency and accuracy of identifying potential mRNA-drug interactions.

Main Methods:

• HGCLMDA integrates graph convolutional networks and hypergraph convolutional networks to capture high-order relationships.
• A cross-view contrastive learning architecture enhances the model's learning capacity.
• Inner product is utilized to calculate mRNA-drug sensitivity association scores.

Main Results:

• HGCLMDA demonstrated superior performance compared to traditional GCN-based methods, contrastive learning methods, and state-of-the-art approaches.
• Visualization experiments confirmed the effectiveness of learned mRNA and drug embeddings.
• Experiments on sparse datasets highlighted the method's robustness and performance.

Conclusions:

• Hypergraph structures play a critical role in enhancing model performance for association prediction.
• HGCLMDA effectively models similarities in mRNA-mRNA and drug-drug interactions.
• The developed method shows significant potential as a tool for predicting mRNA-drug sensitivity associations.