Data-Driven Collective Variables for Enhanced Sampling

Abstract

Designing an appropriate set of collective variables is crucial to the success of several enhanced sampling methods. Here we focus on how to obtain such variables from information limited to the metastable states. We characterize the states by a large set of descriptors and employ neural networks to compress this information on a lower-dimensional space, using Fisher's linear discriminant as objective function to maximize the discriminative power of the network. We benchmark this method on alanine dipeptide, using the non-linearly separable dataset composed by atomic distances. We then study an intermolecular aldol reaction characterized by a concerted mechanism. The resulting variables are able to promote sampling by drawing non-linear paths in the physical space connecting the fluctuations between metastable basins. Lastly, we inspect the behavior of the neural network by studying its relation to physical variables. Through the identification of its most relevant features, we are able to gain chemical insight into the process.

The role of water in host-guest interaction

“…In order to succeed in our endeavour, we rely on a combination of physical considerations and modern machine learning (ML) techniques. In particular, we use a method that we have recently developed that goes under the name of Deep Linear Discriminant Analysis (Deep-LDA) [24]. Deep-LDA builds efficient…”

“…In this work, we are mainly interested in computing the free energy difference ∆G between the bound state (B) in which the ligand sits in the lowest free energy binding pose and the unbound state (U) where the ligand is solvated in water and free to diffuse. In order to obtain a CV able to capture water behaviour we use the recently developed machine learning Deep-LDA method [24].…”

Expanding the boundaries of ligand–target modeling by exascale calculations

“…This represents an additional level of parallelism (Section 2.2). Machine learning (ML) can further boost HPC-based free-energy calculations either by refining existing approaches 89 or by introducing new ones. [90][91][92][93] A hybrid QM/MM interface.…”

Identification of synthesisable crystalline phases of water – a prototype for the challenges of computational materials design

“…With an understanding of thermodynamic stability, one may now consider establishing possible synthesis pathways, using approaches such as forward flux sampling [142], or via the identification of suitable collective variables [143,144].…”