Data-Driven Path Collective Variables

Abstract

Identifying optimal collective variables to model transformations using atomicscale simulations is a long-standing challenge. We propose a new method for the generation, optimization, and comparison of collective variables that can be thought of as a data-driven generalization of the path collective variable concept. It consists of a kernel ridge regression of the committor probability, which encodes a transformation's progress. The resulting collective variable is one-dimensional, interpretable, and differentiable, making it appropriate for enhanced sampling simulations requiring biasing. We demonstrate the validity of the method on two different applications: a precipitation model and the association of Li + and F − in water. For the former, we show that global descriptors such as the permutation invariant vector allow reaching an accuracy far from the one achieved via simpler, more intuitive variables. For the latter, we show that information correlated with the transformation mechanism is contained in the first solvation shell only and that inertial effects prevent the derivation of optimal collective variables from the atomic positions only.

Combining Transition Path Sampling with Data-Driven Collective Variables through a Reactivity-Biased Shooting Algorithm

“…In addition, some approaches combine aspects of the two families, such as collective path variables. , These variables are built to describe the pathways connecting the states, but they are typically used with biased sampling algorithms. Recently, machine learning-based approaches have also been used to construct them in a data-driven manner. , …”

Learning Protein–Ligand Unbinding Pathways via Single-Parameter Community Detection

“…A common problem when searching for paths via trajectory clustering is the need for significant user input. ,− To alleviate potential user input bias, machine learning approaches have been developed that learn RCs on the fly during biasing − (see ref for a recent review). In our case, we want to use a machine learning algorithm that (a) operates unsupervised, as no ground truth is available for training, (b) is fast to screen as many alternative path cluster scenarios as possible, and (c) allows for easy backtracking to refine path RCs from the initial guess of input features.…”

Learning Collective Variables with Synthetic Data Augmentation through Physics-Inspired Geodesic Interpolation

“…While classifier-based methods only utilize the state labels of each data point, in certain cases, we can obtain more informative labels that correlate with the reaction progress, enabling us to train a regressor model instead. Recently, France-Lanord et al introduced a regression approach based on committor probabilities and Lazzeri et al derived an algorithm for the unbiased Boltzmann weights from path sampling simulations. Additionally, Kang et al devised a way to learn the committor on the fly during iterations of enhanced sampling simulations.…”