AlloEF: An Ensemble Model for Protein Allosteric Site Identification Based on Transfer Entropy and Energetic...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Drug Discovery

Background:

• Allostery is essential for protein function regulation.
• Identifying allosteric sites is key for understanding protein mechanisms and developing allosteric drugs.

Purpose of the Study:

• To develop an effective computational method, AlloEF, for predicting protein allosteric sites.
• To improve the accuracy of allosteric site identification beyond traditional pocket regions.

Main Methods:

• Developed AlloEF, a soft-voting classifier integrating LightGBM, Random Forest, and XGBoost.
• Integrated multidimensional features: dynamic properties (transfer entropy), energetic frustration, sequence, structure, and network topology.
• Proposed Spatial Neighborhood Feature Aggregation (SNFA) for residue environmental information.
• Utilized Boruta algorithm for feature selection and SVM-SMOTE for class imbalance.

Main Results:

• AlloEF achieved high performance with an F1 score of 0.630 and MCC of 0.609 on an independent test set.
• The model outperformed existing allosteric site prediction methods.
• AlloEF successfully identified allosteric sites within and beyond canonical allosteric pockets.

Conclusions:

• AlloEF establishes a new benchmark for allosteric site prediction.
• The method enhances understanding of protein allosteric mechanisms.
• AlloEF is valuable for the rational design of allosteric modulators.