State predictive information bottleneck

Abstract

The ability to make sense of the massive amounts of high-dimensional data generated from molecular dynamics simulations is heavily dependent on the knowledge of a low-dimensional manifold (parameterized by a reaction coordinate or RC) that typically distinguishes between relevant metastable states, and which captures the relevant slow dynamics of interest. Methods based on machine learning and artificial intelligence have been proposed over the years to deal with learning such low-dimensional manifolds, but they are often criticized for a disconnect from more traditional and physically interpretable approaches. To deal with such concerns, in this work we propose a deep learning based state predictive information bottleneck approach to learn the RC from high-dimensional molecular simulation trajectories. We demonstrate analytically and numerically how the RC learnt in this approach is connected to the committor in chemical physics and can be used to accurately identify transition states. A crucial hyperparameter in this approach is the time delay or how far into the future the algorithm should make predictions about. Through careful comparisons for benchmark systems, we demonstrate that this hyperparameter choice gives useful control over how coarse-grained we want the metastable state classification of the system to be. We thus believe that this work represents a step forward in systematic application of deep learning based ideas to molecular simulations.

Accelerating all-atom simulations and gaining mechanistic understanding of biophysical systems through State Predictive Information Bottleneck

“…As shown in Ref. 16 such a state predictive information bottleneck approximates the perfect RC as given by the committor. 18 In this work, we demonstrate how the RC as learnt through SPIB performed on short, under-sampled trajectories can be used as a biasing coordinate in enhanced sampling, allowing significant and nearly automated acceleration of protein conformational dynamics and small molecule permeation through biological membranes.…”

“…In this regard, SPIB can be thought of as a 'fast mode filter' where the hyperparameter, ∆t can be used to tune the coarse-graining of the identified slow modes, as demonstrated in its original proof-of-principle publication. 16 Thus, for a given unbiased trajectory {X 1 , • • • , X M +s } and its corresponding state labels {y 1 , • • • , y M +s } with large enough M , we can employ the deep variational information bottleneck framework 16,29 and construct an artificial neural network (ANN) that is trained to maximize the following objective function:…”

“…This challenge clearly illustrates the need for developing novel algorithms to interpret MD results to learn the RC and tackle the rare event problem. State Predictive Information Bottleneck (SPIB) 16 is one such recent method that enables learning the RC in systems with an arbitrary and a priori unknown number of metastable states. It belongs to the Reweighted Autoencoded Variational Bayes (RAVE) 17 family of methods.…”

Making high-dimensional molecular distribution functions tractable through Belief Propagation on Factor Graphs

“…We conclude by noting that despite the limitations of our simple illustrative strategies, this work can also be combined with sampling methods that are used to determine effective sets of basis functions for describing molecular systems 43 or those that are used for dimensionality reduction. 36,39,44 We envision that these methods will be used to develop an automated framework for sampling biomolecules that takes a starting structure and runs enhanced sampling simulations without the necessity for expert knowledge. This combination should allow simulation of new systems where traditional lowdimensional biasing variables are unsuccessful in a highthroughput manner, for instance biomolecular complexes with more than one constituent.…”

Interrogating RNA-small molecule interactions with structure probing and AI augmented-molecular simulations

“…The RC itself is learned as the most informative low-dimensional representation expressed as a past-future information bottleneck. 27 By efficiently sampling multiple independent riboswitch-ligand dissociation events for both cognate and synthetic ligands, we are able to reproduce binding affinity and flexibility measurements from MST, FIA and SHAPE-MaP experiments. Furthermore, we are able to pin-point which nucleotides play the most critical roles in dissociation.…”