Prediction of Protein Allosteric Sites with Transfer Entropy and Spatial Neighbor-Based Evolutionary Information...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Drug Discovery

Background:

• Allosteric regulation is crucial for protein function, offering advantages over orthosteric drug targeting due to diverse allosteric sites.
• Developing selective allosteric modulators requires accurate identification of these regulatory sites.
• Existing methods for allosteric site prediction face challenges in feature representation and class imbalance.

Purpose of the Study:

• To develop a robust and accurate computational method for predicting protein allosteric sites.
• To integrate novel features, including dynamic properties and evolutionary information, into allosteric site prediction.
• To address the class imbalance problem inherent in predicting rare allosteric sites.

Main Methods:

• Developed an ensemble machine learning model named AllosES.
• Utilized novel features: entropy transfer-based dynamic properties, secondary structure, spatial neighbor-based evolutionary information, and physicochemical properties.
• Implemented a multiple grouping strategy to handle class imbalance during feature selection and model construction.

Main Results:

• AllosES achieved a Matthews Correlation Coefficient (MCC) of 0.556 on the independent D24 test set.
• Successfully ranked true allosteric sites within the top three predictions for 83.3% (D24) and 89.3% (D28) of proteins.
• Outperformed existing state-of-the-art methods in allosteric site prediction accuracy.

Conclusions:

• AllosES demonstrates significant promise as a computational tool for identifying protein allosteric sites.
• The integration of advanced features and ensemble learning effectively addresses prediction challenges.
• This method can facilitate the design of novel allosteric modulators for therapeutic applications.