AlloBench: A Data Set Pipeline for the Development and Benchmarking of Allosteric Site Prediction Tools

Area of Science:

• Biochemistry and Structural Biology
• Computational Biology
• Bioinformatics

Background:

• Allostery is crucial for regulating macromolecular activity via distant effector binding.
• Existing allosteric datasets are outdated and unsuitable for data-intensive computational studies.
• There is a need for comprehensive, up-to-date resources for studying protein allostery.

Purpose of the Study:

• To present the AlloBench pipeline for generating high-quality biomolecular datasets with allosteric and active site information.
• To create a robust dataset suitable for computational and data-driven research on protein allostery.
• To benchmark existing allosteric site prediction tools on a standardized dataset.

Main Methods:

• Integrated data from multiple databases (AlloSteric Database, UniProt, MCSA, PDB) using the AlloBench pipeline.
• Generated a dataset of 2141 allosteric sites from 2034 protein structures.
• Evaluated seven allosteric site prediction tools (APOP, PASSer, Ohm, ALLO, Allosite, STRESS, AlloPred) on a subset of 100 proteins.

Main Results:

• The AlloBench pipeline successfully created a large, high-quality dataset for studying allostery.
• All evaluated allosteric site prediction tools demonstrated accuracy below 60%.
• PASSer (Ensemble) showed superior performance compared to other tested prediction tools.

Conclusions:

• The AlloBench dataset provides a valuable resource for advancing the study of protein allostery.
• Current allosteric site prediction tools require significant improvement.
• AlloBench will facilitate the development of more accurate prediction tools and serve as a general reference for allostery research.