SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning is crucial for efficient disease treatment.
• Existing computational methods for drug-disease prediction often overlook non-Euclidean data and multisource information.
• Graph neural networks face challenges in optimizing feature diffusion distance.

Purpose of the Study:

• To propose SiSGC, a novel computational model for drug-disease association prediction.
• To leverage biological knowledge and heterogeneous graph structures for improved prediction accuracy.
• To address limitations in existing graph neural network approaches for feature diffusion.

Main Methods:

• SiSGC utilizes biological knowledge as initial features and learns structural information from a heterogeneous graph.
• The model adaptively selects information diffusion distance and fuses structural features with denoised similarity information.
• Predictions are made using the CatBoost classifier.

Main Results:

• SiSGC demonstrates superior performance over six leading methods and four variants across three datasets and two splitting strategies.
• The model's robustness and generalization capabilities were confirmed.
• A case study on breast neoplasms validated SiSGC's trustworthiness and simplicity.

Conclusions:

• SiSGC offers a robust and effective approach to drug repositioning by integrating diverse data sources and advanced graph learning techniques.
• The model successfully identified four potential drugs for breast cancer treatment with high confidence.
• SiSGC serves as a valuable tool for accelerating drug discovery and development.