A Hybrid OPES-eABF Framework for Efficient Exploration and Data-Driven Collective Variable Discovery in Complex...

Area of Science:

• Computational chemistry and biophysics
• Molecular dynamics simulations
• Machine learning in science

Background:

• Molecular dynamics (MD) simulations are limited by accessible timescales for rare events.
• Enhanced sampling methods accelerate rare events but require predefined collective variables (CVs).
• Automated discovery of relevant CVs is crucial for efficient simulations.

Purpose of the Study:

• To develop and validate a unified computational framework for enhanced biomolecular simulations.
• To integrate a hybrid enhanced sampling scheme (OPES-eABF) with machine learning for automated CV discovery.
• To accurately predict rare events like protein folding and ligand unbinding.

Main Methods:

• Coupling the OPES-eABF hybrid scheme with a machine learning workflow (Koopman-reweighted DeepTICA-LASSO).
• Testing the framework on alanine dipeptide, chignolin folding, and benzene unbinding from T4 lysozyme.
• Analyzing the FULLSAMPLES/PACE ratio for optimal sampling efficiency and ergodicity.

Main Results:

• The OPES-eABF hybrid method demonstrated faster sampling than individual methods across tested systems.
• Machine-learned CVs successfully captured slow dynamics and generated physically meaningful free-energy landscapes.
• Accurate prediction of benzene unbinding free energy from T4 lysozyme, closely matching experimental values.

Conclusions:

• The integrated framework offers a robust, transferable, and interpretable approach for rare-event simulations.
• Data-driven CV optimization significantly enhances the efficiency and accuracy of enhanced sampling methods.
• This approach advances the study of complex kinetics and thermodynamics in biomolecular systems.