Predictive collective variable discovery with deep Bayesian models

Abstract

Extending spatio-temporal scale limitations of models for complex atomistic systems considered in biochemistry and materials science necessitates the development of enhanced sampling methods. The potential acceleration in exploring the configurational space by enhanced sampling methods depends on the choice of collective variables (CVs). In this work, we formulate the discovery of CVs as a Bayesian inference problem and consider the CVs as hidden generators of the full-atomistic trajectory. The ability to generate samples of the fine-scale atomistic configurations using limited training data allows us to compute estimates of observables as well as our probabilistic confidence on them. The methodology is based on emerging methodological advances in machine learning and variational inference. The discovered CVs are related to physicochemical properties which are essential for understanding mechanisms especially in unexplored complex systems. We provide a quantitative assessment of the CVs in terms of their predictive ability for alanine dipeptide (ALA-2) and ALA-15 peptide.

Biomolecular modeling thrives in the age of technology

“…Enhanced sampling techniques are now being combined with machine learning to improve the selection of collective variables 108 and to develop new methods 109,110 . Clearly, artificial intelligence and ML algorithms are changing the way we do molecular modeling.…”

Collective variable-based enhanced sampling and machine learning

“… is another normal distribution with mean generated by the decoder neural network and standard deviation as hyperparameters. The decoder can be easily adopted as a configuration generator [ 95 ], while applying the encoder as CVs is not straightforward. Variational autoencoder suggests that is a random variable instead of a deterministic function of .…”

“…Although there are very limited studies on sampling all-atom configurations with machine learning methods, these developments are changing the world of all-atom sampling [ 95 , 111 , 173 – 176 ]. Here we will briefly discuss some methods as interesting directions.…”

Computational methods for exploring protein conformations

“…Neural networks [73], including nonlinear [50,74,75] and time-lagged variational [76] autoencoders, and a Bayesian framework that operates according to similar principles [77], have also been developed to find the optimal slow CVs. In some cases (e.g.…”