GCMM: graph convolution network based on multimodal attention mechanism for drug repurposing.

Area of Science:

• Computational drug discovery
• Artificial intelligence in medicine
• Pharmacology

Background:

• Predicting drug-disease relationships is crucial for in silico drug repurposing.
• Current computational models struggle to reliably predict these associations from diverse data sources.

Purpose of the Study:

• To introduce a novel end-to-end model, Graph convolution network based on a multimodal attention mechanism (GCMM), for enhanced drug-disease association prediction.
• To improve the reliability and accuracy of identifying potential drug repurposing candidates.

Main Methods:

• GCMM integrates known drug-disease relations, drug-drug similarities (chemical, therapeutic), and disease-disease similarities (semantic, target-based) into a heterogeneous network.
• A Graph Convolution Network encoder learns embeddings for diseases and drugs.
• A multimodal attention layer assigns differential importance to various features and multi-source information.

Main Results:

• GCMM demonstrated superior performance compared to four recent deep learning models in 5-fold cross-validation.
• The model reliably predicts drug-disease relationships, showing significant improvements in key metrics.
• A case study on Alzheimer's disease validated GCMM's predictions, with four of five top candidates supported by existing literature.

Conclusions:

• GCMM offers a robust and effective tool for drug development and repositioning.
• The model's architecture, particularly the multimodal attention mechanism, is essential for its high predictive performance.
• Further research can leverage GCMM for accelerated identification of novel therapeutic applications for existing drugs.