Explainable Machine Learning Guided Enhanced Sampling of Protein Conformational Transition in HSP90.

Area of Science:

• Computational Biology
• Biophysics
• Molecular Dynamics Simulations

Background:

• Understanding protein conformational transitions is crucial for molecular dynamics simulations.
• Heat shock protein 90 (HSP90) plays a vital role in cellular processes, and its conformational changes are key to its function.
• Traditional simulation methods face challenges in exploring complex free-energy landscapes of slow transitions.

Purpose of the Study:

• To elucidate the thermodynamics, kinetics, and mechanisms of the ATP-lid conformational transition in HSP90.
• To develop and apply explainable machine learning (ML)-based collective variables (CVs) for enhanced sampling.
• To provide atomistic mechanistic insights and inform future inhibitor design for HSP90.

Main Methods:

• Employed enhanced sampling simulations utilizing explainable ML-based CVs.
• Explored the free-energy landscape of the millisecond-time scale ATP-lid conformational transition in HSP90.
• Integrated biased enhanced sampling with an unbiased weighted ensemble algorithm for kinetic rate determination.

Main Results:

• Achieved relative free energies consistent with nuclear magnetic resonance (NMR) experiments at reduced computational cost.
• Identified key residues involved in the ATP-lid transition through interpretable ML-CVs, offering mechanistic insights.
• Demonstrated transferability of the ML-CV to mutant variants, accurately reproducing experimental population shifts.

Conclusions:

• The developed framework provides detailed thermodynamic, kinetic, and mechanistic insights into HSP90 conformational transitions.
• Explainable ML-based CVs offer a powerful approach for investigating complex biomolecular conformational landscapes.
• This methodology has significant potential for informing future inhibitor design strategies targeting HSP90.