PASSer: Prediction of Allosteric Sites Server

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Allosteric regulation is vital for protein function and a key target in drug development.
• Identifying allosteric sites is a critical step for rational drug design.
• Existing computational methods for allosteric site prediction often rely on specific features or dynamics.

Purpose of the Study:

• To develop and validate a novel ensemble learning method for predicting protein allosteric sites.
• To leverage the strengths of eXtreme gradient boosting (XGBoost) and graph convolutional neural networks (GCNN) for this task.
• To provide a computational tool that aids in identifying potential allosteric targets for drug discovery.

Main Methods:

• An ensemble learning model integrating XGBoost and GCNN was developed.
• The model learns physical properties and topological information without prior assumptions.
• Performance was evaluated using multiple metrics on a test set of protein allosteric sites.

Main Results:

• The ensemble model demonstrated strong performance in predicting allosteric sites.
• 84.9% of known allosteric pockets were identified within the top 3 predictions.
• The method effectively captures relevant protein features for site identification.

Conclusions:

• The developed XGBoost and GCNN ensemble model is effective for predicting protein allosteric sites.
• This computational approach offers a valuable tool for accelerating drug discovery and design.
• The PASSer server and CLI provide accessible resources for researchers.