How Well Can AI and Physics-Based Simulations Predict the Probability a Cryptic Pocket Is Open?

Area of Science:

• Computational biology
• Drug discovery
• Artificial intelligence in biochemistry

Background:

• Artificial intelligence (AI) models are advancing protein structure prediction and functional annotation.
• Molecular dynamics (MD)-inspired generative models like AlphaFlow and BioEmu excel at capturing protein conformational ensembles.
• Cryptic pockets are transient protein sites vital for drug discovery, offering new therapeutic targets.

Purpose of the Study:

• To benchmark AI-driven generative models (AlphaFlow, BioEmu) and residue-level predictors (PocketMiner, CryptoBank) against physics-based MD simulations.
• To evaluate the capability of these computational methods in detecting cryptic pockets in proteins.
• To assess the performance of these methods in capturing the effects of mutations on cryptic pocket formation.

Main Methods:

• Benchmarking AI generative models (AlphaFlow, BioEmu) and residue-level predictors (PocketMiner, CryptoBank).
• Utilizing physics-based molecular dynamics (MD) simulations as a comparison.
• Testing methods on interferon inhibitory domain of Zaire Ebola VP35 (VP35) and TEM-1 β-lactamase (TEM) proteins and their mutants.

Main Results:

• All tested methods successfully identified pockets in VP35 and differentiated between mutants that opened or closed pockets.
• Performance was inconsistent for TEM-1 β-lactamase, particularly where pocket opening was subtle.
• AI and simulation-based strategies show potential but require further refinement for robust, system-independent cryptic pocket detection.

Conclusions:

• AI and simulation approaches hold significant promise for identifying cryptic pockets in drug discovery.
• Current methods demonstrate varying efficacy, especially with subtle conformational changes.
• Further research is needed to enhance the robustness and generalizability of these computational tools for cryptic pocket prediction.