Hierarchical AF2RAVE for Multiconformation Virtual Screening Targeting S100 Ca²⁺-Binding Proteins.

Area of Science:

• Computational Biology
• Structural Biology
• Drug Discovery

Background:

• Protein function relies on dynamic conformational changes, presenting challenges for drug design.
• Modeling protein backbone and side chain metastability is crucial for accurate structure-based drug discovery.
• Generating diverse protein conformations for virtual screening is computationally demanding.

Purpose of the Study:

• To introduce AF2RAVE, a hierarchical pipeline integrating AlphaFold2 with enhanced sampling for protein structure modeling.
• To systematically explore the free energy landscape and metastability of proteins at backbone and side chain levels.
• To enable the generation of diverse, holo-like protein conformations from sequence for drug discovery applications.

Main Methods:

• Integration of AlphaFold2 with machine learning-based enhanced sampling (AF2RAVE pipeline).
• Systematic exploration of protein free energy landscapes and metastability.
• Application to the calcium-binding S100 protein family to generate diverse conformations.
• Retrospective docking and enrichment testing using a novel Ca2+-S100B inhibitor dataset.

Main Results:

• AF2RAVE successfully generated diverse holo-like conformations for the S100 protein family from sequence.
• AF2RAVE-generated structures demonstrated superior performance in retrospective docking and enrichment testing compared to standard AlphaFold2.
• The method outperformed experimentally resolved X-ray structures in enrichment testing for Ca2+-S100B inhibitors.

Conclusions:

• AF2RAVE offers a powerful approach for modeling protein metastability and conformational dynamics.
• The pipeline significantly enhances high-throughput virtual screening capabilities.
• AF2RAVE shows great potential for selective inhibitor discovery, especially for challenging targets like the S100 protein family.