Graph Convolutional Autoencoder and Fully-Connected Autoencoder with Attention Mechanism Based Method for Predicting...

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repurposing accelerates development and reduces costs by identifying new uses for existing drugs.
• Previous methods often fail to fully integrate multiple drug similarities and exploit network topology.

Purpose of the Study:

• To propose GFPred, a novel method for predicting drug-disease associations by deeply integrating diverse data sources.
• To leverage graph convolutional autoencoders and attention mechanisms for enhanced prediction accuracy.

Main Methods:

• GFPredict integrates drug-disease associations, disease similarities, three drug similarities, and node attributes.
• It constructs heterogeneous networks and employs graph convolutional autoencoders with an attribute-level attention mechanism.
• A fully-connected autoencoder and a convolutional neural network combine topology, attribute, and feature representations.

Main Results:

• Ablation studies confirm the contribution of different drug attributes.
• GFPredict outperforms existing state-of-the-art methods in predicting drug-disease associations.
• Case studies demonstrate GFPred's ability to identify actual drug-disease associations.

Conclusions:

• GFPredict offers a powerful and effective approach for drug repurposing.
• The method facilitates the discovery of novel drug-disease associations for experimental validation.