Abstract:We consider the problem of sampling transition paths between two given metastable states of a molecular system, e.g. a folded and unfolded protein or products and reactants of a chemical reaction. Due to the existence of high energy barriers separating the states, these transition paths are unlikely to be sampled with standard Molecular Dynamics (MD) simulation. Traditional methods to augment MD with a bias potential to increase the probability of the transition rely on a dimensionality reduction step based on Collective Variables (CVs). Unfortunately, selecting appropriate CVs requires chemical intuition and traditional methods are therefore not always applicable to larger systems. Additionally, when incorrect CVs are used, the bias potential might not be minimal and bias the system along dimensions irrelevant to the transition. Showing a formal relation between the problem of sampling molecular transition paths, the Schrödinger bridge problem and stochastic optimal control with neural network policies, we propose a machine learning method for sampling said transitions. Unlike previous non-machine learning approaches our method, named PIPS, does not depend on CVs. We show that our method successful generates low energy transitions for Alanine Dipeptide as well as the larger Polyproline and Chignolin proteins.

What problem does this paper attempt to address?

The paper aims to address the problem of sampling transition paths between two metastable states in molecular systems, particularly for high energy barrier issues encountered in processes such as protein folding, conformational changes, and chemical reactions. Traditional molecular dynamics (MD) simulations often struggle to effectively sample these transition paths because the energy barriers they need to cross are usually very high. To tackle this challenge, the researchers propose a method called PIPS (Path Integral Path Sampling), which is a new approach based on path integral stochastic optimal control theory. Unlike traditional methods that rely on Collective Variables (CVs), PIPS does not depend on predefined CVs but operates directly on the overall geometric structure of the molecule. The advantage of this method is that it can be applied to large systems where finding suitable CVs is difficult. Specifically, PIPS uses stochastic optimal control theory to train a parameterized bias potential to improve the efficiency of sampling transition paths from one metastable state to another. This method not only overcomes the difficulty of selecting appropriate CVs but can also be effectively applied to larger molecular systems. The paper experimentally validates the method through three different molecular systems: Alanine Dipeptide, Polyproline, and Chignolin protein. The experimental results show that the PIPS method can successfully generate low-energy transition paths without relying on CVs and has higher efficiency and accuracy compared to traditional methods. Additionally, PIPS demonstrates the ability to identify the correct collective variables for sampling even when the target state is not specified correctly. This indicates that PIPS could be used to validate the effectiveness of candidate collective variables.

Stochastic Optimal Control for Collective Variable Free Sampling of Molecular Transition Paths

Enhanced Sampling with Sub-optimal Collective Variables: Reconciling Accuracy and Convergence Speed

Generative methods for sampling transition paths in molecular dynamics

Transition Path Sampling with Boltzmann Generator-based MCMC Moves

Transition Path Sampling with Improved Off-Policy Training of Diffusion Path Samplers

Sampling a rare protein transition with a hybrid classical-quantum computing algorithm

Acceleration with Interpretability: Surrogate Model Based Collective Variable for Enhanced Sampling

Adaptive CVgen: Leveraging Reinforcement Learning for Advanced Sampling in Protein Folding and Chemical Reactions

Data-Driven Collective Variables for Enhanced Sampling

Adaptive Sampling Methods for Molecular Dynamics in the Era of Machine Learning

Combining transition path sampling with data-driven collective variables through a reactivity-biased shooting algorithm

Enhanced Sampling of Biomolecular Slow Conformational Transitions Using Adaptive Sampling and Machine Learning

TRANSITION PATH THEORY FOR LANGEVIN DYNAMICS ON MANIFOLDS: OPTIMAL CONTROL AND DATA-DRIVEN SOLVER

Augmenting Control over Exploration Space in Molecular Dynamics Simulators to Streamline De Novo Analysis through Generative Control Policies

Sampling a Rare Protein Transition Using Quantum Annealing

Doob's Lagrangian: A Sample-Efficient Variational Approach to Transition Path Sampling

Collective Variable-Based Enhanced Sampling: from Human Learning to Machine Learning

Enhanced sampling of robust molecular datasets with uncertainty-based collective variables

Sampling metastable systems using collective variables and Jarzynski-Crooks paths

Neural Network and Nearest Neighbor Algorithms for Enhancing Sampling of Molecular Dynamics

Sampling in In Silico Biomolecular Studies: Single-Stage Experiments vs Multiscale Approaches.