Multi-view semi-supervised adversarial attention network for drug repurposing.

Area of Science:

• Computational drug discovery
• Bioinformatics
• Artificial intelligence in medicine

Background:

• Drug repositioning offers an efficient route to identify new therapeutic uses for existing drugs.
• Current prediction methods often lack biological and chemical insights, and suffer from data sparsity and poor negative sample selection.
• These limitations hinder the accuracy and reliability of drug-disease association predictions.

Purpose of the Study:

• To develop an advanced computational model for predicting drug-disease associations.
• To overcome the limitations of existing methods, including data sparsity and reliance on similarity data.
• To enhance the robustness and accuracy of drug repositioning strategies.

Main Methods:

• Proposed the Multi-view Semi-Supervised Adversarial Attention Network (M-SSAAN).
• Integrated structural and similarity embeddings for multi-view representations.
• Utilized attention mechanisms and semi-supervised adversarial learning to address data sparsity and improve model robustness.

Main Results:

• M-SSAAN demonstrated effectiveness through comparative experiments and ablation studies.
• Case studies, including ibuprofen indications and zero-shot lung cancer predictions, validated the model's predictive capabilities.
• The model successfully identified reliable drug-disease associations, showcasing its practical utility.

Conclusions:

• M-SSAAN shows significant potential as a robust tool for drug discovery.
• The developed network effectively predicts drug-disease associations, aiding in the identification of new therapeutic applications.
• This approach advances the field of computational drug repositioning.