Developing a machine learning model to identify protein-protein interaction hotspots to facilitate drug discovery

Area of Science:

• Computational biology
• Drug discovery
• Machine learning in medicine

Background:

• Traditional drug discovery often relies on enzymatic targets.
• Protein-protein interfaces are emerging targets for diseases like cancer and HIV.
• Current computational models for interface hotspot prediction have limited accuracy (~70%) and often neglect structural features.

Purpose of the Study:

• To develop and evaluate a machine learning model for enhanced prediction of protein-protein interface hotspots.
• To integrate diverse features, including amino acid chain information, into the predictive model.
• To perform virtual drug screening on identified hotspots for potential therapeutic applications.

Main Methods:

• Development of a novel machine learning model incorporating multi-feature integration.
• Evaluation of the model's predictive performance using AUROC, sensitivity, and specificity.
• Virtual screening of predicted hotspots on the EphB2-ephrinB2 complex.

Main Results:

• The machine learning model achieved high predictive accuracy with an AUROC of 0.842, sensitivity of 0.833, and specificity of 0.850.
• Virtual screening identified potential drug candidates for diseases associated with EphB2-ephrinB2 overexpression, including various cancers.
• The model successfully predicted known drug-disease associations and identified nadolol as a potential therapeutic agent for multiple cancers.

Conclusions:

• The developed machine learning model significantly improves the prediction of protein-protein interface hotspots.
• This approach facilitates the identification of novel drug targets and potential therapeutic agents for complex diseases.
• The study highlights the potential of repurposing existing drugs, such as beta-blockers, for cancer treatment.