CDPMF-DDA: contrastive deep probabilistic matrix factorization for drug-disease association prediction.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Drug development is complex and costly.
• Drug-disease association (DDA) prediction identifies new uses for existing drugs.
• Existing single-view contrastive learning methods have limitations in capturing complex drug-disease relationships.

Purpose of the Study:

• To introduce CDPMF-DDA, a novel multi-view contrastive learning framework for enhanced drug-disease association prediction.
• To improve the accuracy and robustness of identifying new therapeutic uses for existing medications.
• To leverage diverse information representations for better understanding of drug-disease interactions.

Main Methods:

• Decomposed drug-disease association matrix into drug and disease feature matrices.
• Reconstructed drug-disease, drug-drug, and disease-disease similarity networks to reduce noise.
• Generated multiple contrastive views from original and reconstructed networks to capture hidden feature associations.

Main Results:

• CDPMF-DDA achieved an average AUC of 0.9475 and AUPR of 0.5009 on three standard datasets, outperforming existing models.
• Case studies on Alzheimer's disease and epilepsy validated the model's effectiveness and robustness.
• The multi-view approach effectively captured complex drug-disease associations.

Conclusions:

• CDPMF-DDA, a multi-view contrastive learning framework, effectively integrates multi-source information for DDA prediction.
• The model demonstrates high accuracy and robustness, making it a powerful tool for drug repositioning.
• This framework advances the discovery of new therapeutic strategies and drug repurposing.