GraphCPIs: A novel graph-based computational model for potential compound-protein interactions.

Area of Science:

• Computational drug discovery
• Bioinformatics
• Cheminformatics

Background:

• Identifying compound-protein interactions (CPIs) is crucial for drug discovery.
• Traditional CPI prediction methods face challenges and limitations.
• Computer-aided methods offer rapid identification of high-quality CPI candidates.

Purpose of the Study:

• To propose a novel computational model, GraphCPIs, for improving the accuracy of CPI prediction.
• To leverage graph-based features and machine learning for enhanced CPI identification.

Main Methods:

• Constructing an adjacent matrix representing drug and protein entities.
• Employing graph convolutional networks and Grarep embedding for feature representation.
• Utilizing an extreme gradient boosting (XGBoost) classifier with stacked features for prediction.

Main Results:

• GraphCPIs achieved an average predictive accuracy of 90.09%.
• The model demonstrated strong performance with an average AUC-ROC of 0.9572 and AUC-PR of 0.9621.
• Comparative experiments showed GraphCPIs outperformed state-of-the-art methods.

Conclusions:

• The GraphCPIs model significantly enhances CPI prediction accuracy.
• This approach offers a valuable tool for discovering novel drug-related proteins.
• GraphCPIs provides a robust computational strategy for accelerating drug discovery pipelines.