A Computational Framework for Predicting Novel Drug Indications Using Graph Convolutional Network With Contrastive...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Inferring drug indications is crucial but experimentally costly.
• Graph learning methods are emerging for drug indication prediction.
• Limited labeled data poses a challenge for traditional methods.

Purpose of the Study:

• To develop an efficient computational model for predicting novel drug indications.
• To leverage contrastive learning to overcome data scarcity in drug indication prediction.
• To integrate diverse biological data for enhanced drug-disease association inference.

Main Methods:

• Developed DrIGCL, a model combining graph convolutional networks and contrastive learning.
• Incorporated drug structure, disease comorbidities, and known drug indications for feature extraction.
• Utilized a hybrid loss function combining contrastive and classification objectives.

Main Results:

• DrIGCL consistently outperformed existing computational methods in drug indication prediction, especially for top-k predictions.
• Ablation studies confirmed the significant contribution of contrastive learning to predictive performance.
• The model's practical utility was demonstrated through the prediction of novel indications for Aspirin.

Conclusions:

• DrIGCL offers a powerful and data-efficient approach for predicting drug indications.
• The integration of graph learning and contrastive learning advances computational drug discovery.
• The model shows promise for accelerating the identification of new therapeutic uses for existing drugs.