iSEE: Interface structure, evolution, and energy-based machine learning predictor of binding affinity changes upon...

Area of Science:

• Computational biology
• Protein engineering
• Drug design

Background:

• Accurate prediction of binding affinity changes upon mutation is vital for protein engineering and drug design.
• Machine learning (ML) methods are increasingly used, but require sensitive features due to limited experimental data for robustness.

Purpose of the Study:

• To develop a fast and reliable predictor of binding affinity changes upon single point mutation.
• To utilize a limited set of Structure, Evolution, and Energy-based features for accurate predictions.

Main Methods:

• Developed iSEE, a random forest-based prediction method.
• Employed a limited set of 31 interface Structure, Evolution, and Energy-based features.
• Trained and validated on a diverse dataset of 1102 mutations in 57 protein-protein complexes.

Main Results:

• iSEE achieved a high prediction performance (PCC of 0.80, RMSE of 1.41 kcal/mol) on the training dataset.
• The method demonstrated competitive performance against state-of-the-art approaches on blind test datasets.
• Feature analysis highlighted the importance of evolutionary conservation for predicting mutation effects.

Conclusions:

• iSEE offers a robust and efficient approach for predicting mutation-induced binding affinity changes.
• The method's performance suggests the utility of Structure, Evolution, and Energy features.
• Demonstrated the application of iSEE through mutation scanning of the MDM2-p53 complex interface.