Coupling dynamics and evolutionary information with structure to identify protein regulatory and functional binding...

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Proteins possess functional (active) and regulatory (allosteric) binding sites critical for their activity.
• Identifying these binding sites is essential for understanding protein function and disease.
• Existing methods often struggle to accurately predict both types of binding sites.

Purpose of the Study:

• To develop a novel computational method, Active and Regulatory site Prediction (AR-Pred), for identifying protein active and allosteric sites.
• To integrate protein dynamics with traditional features for enhanced binding site prediction.
• To evaluate the performance of AR-Pred against existing prediction tools.

Main Methods:

• AR-Pred utilizes a machine learning approach (random forest) trained on diverse datasets.
• The method incorporates protein geometry, evolutionary information, physicochemical properties, and intrinsic protein dynamics.
• Ensemble models were generated for both active and allosteric site predictions.

Main Results:

• Active site prediction models achieved a median AUC of 91% and MCC of 0.68.
• Allosteric site prediction models achieved a median AUC of 80% and MCC of 0.48.
• AR-Pred demonstrated competitive or superior performance compared to existing methods on independent test sets.

Conclusions:

• AR-Pred effectively predicts both active and allosteric protein binding sites by integrating protein dynamics.
• The method offers improved accuracy, particularly for the challenging prediction of allosteric sites.
• AR-Pred is available as a free, downloadable package, facilitating its use in biological research.