Designing transferable transition state guided collective variables via interpretable machine learning models for...

Area of Science:

• Computational Biophysics
• Molecular Dynamics Simulations
• Machine Learning in Chemistry

Background:

• Enhanced sampling methods are crucial for studying complex biophysical processes at atomistic resolution, overcoming timescale limitations in molecular dynamics (MD) simulations.
• Selecting appropriate collective variables (CVs) for enhanced sampling remains a significant challenge.
• Machine learning (ML) algorithms show promise for designing efficient CVs but often lack interpretability and transferability.

Purpose of the Study:

• To develop a general, interpretable, and transferable enhanced sampling method using surrogate models.
• To address the limitations of ML-based CVs, specifically their interpretability and transferability across different systems.
• To improve the efficiency of free energy surface (FES) exploration in molecular simulations.

Main Methods:

• Introduced a surrogate model-assisted enhanced sampling method utilizing an elastic net (EN) regression model.
• The EN model expresses the relevance of different order parameters (OPs) as a linear combination locally at the transition state (TS) region.
• Applied TS-derived CVs to explore polymer collapse transitions in systems of varying lengths.

Main Results:

• Demonstrated successful application of surrogate model-based TS-derived CVs in exploring polymer collapse transition landscapes.
• Achieved faster free energy convergence within very short simulation times compared to other tested OPs.
• Showcased the transferability of the approach across different polymer lengths without requiring extensive retraining for each system.

Conclusions:

• The developed method provides a general and interpretable approach for enhanced sampling simulations.
• Surrogate model-assisted TS-derived CVs can be effectively extrapolated beyond their training systems.
• This approach offers a promising alternative to existing methods for studying complex biophysical processes.