SVCPI: A Soft Voting Ensemble-Based Model for Compound-Protein Interaction Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Understanding compound-protein interactions (CPI) is vital for drug discovery and development.
• Computational methods, particularly deep learning, are valuable tools for predicting CPI.
• Improving prediction accuracy requires effective feature extraction and model integration strategies.

Purpose of the Study:

• To introduce SVCPI, a soft voting ensemble model for enhanced CPI prediction.
• To evaluate SVCPI's performance against basic classifiers and state-of-the-art methods.
• To demonstrate the efficacy of integrating GCN and molecular fingerprint features for CPI prediction.

Main Methods:

• Developed SVCPI, a soft voting ensemble model.
• Utilized Graph Convolutional Network (GCN) features and molecular fingerprint features.
• Integrated multiple basic classifiers using a soft voting strategy.

Main Results:

• SVCPI outperformed individual basic classifiers across diverse datasets.
• SVCPI demonstrated competitive performance against classical machine learning and state-of-the-art approaches.
• SVCPI achieved significant improvements in AUC-ROC (>10%) and AUC-PR (>20%) on the Kinases dataset compared to leading methods.

Conclusions:

• SVCPI is an effective and feasible computational method for compound-protein interaction prediction.
• The soft voting ensemble approach enhances prediction accuracy by integrating diverse features.
• SVCPI shows particular promise for drug discovery applications, especially in kinase-related research.