Accelerating All-Atom Simulations and Gaining Mechanistic Understanding of Biophysical Systems through State Predictive Information Bottleneck

Abstract

An effective implementation of enhanced sampling algorithms for molecular dynamics simulations requires a priori knowledge of the approximate reaction coordinate describing the relevant mechanisms in the system. In this work, we focus on the recently developed artificial intelligence-based State Predictive Information Bottleneck (SPIB) approach and demonstrate how SPIB can learn such a reaction coordinate as a deep neural network even from undersampled trajectories. We exemplify its usefulness by achieving more than 40 times acceleration in simulating two model biophysical systems through well-tempered metadynamics performed by biasing along the SPIB-learned reaction coordinate. These include left-to right-handed chirality transitions in a synthetic helical peptide (Aib) 9 and permeation of a small benzoic acid molecule through a synthetic, symmetric phospholipid bilayer. In addition to significantly accelerating the dynamics and achieving back and forth movement between different metastable states, the SPIB-based reaction coordinate gives mechanistic insights into the processes driving these two important problems.

Quantifying Energetic and Entropic Pathways in Molecular Systems

“…As a more challenging problem, we consider small-molecule permeation through membranes, which is an important process for determining the efficacy of pharmaceuticals as well as other biological processes. , However, it is known that for a variety of small molecules the membrane permeation cannot be adequately described using a single reaction coordinate. , Recently it was shown that one could use SPIB to find an adequate reaction coordinate for enhancing the sampling of benzoic acid (BA) crossing a 1,2-dimyristoyl- sn -glycero-3-phosphocholine (DMPC) membrane bilayer …”

“…Furthermore, SPIB, like its predecessor RAVE (reweighted autoencoded variational Bayes), , is used to iteratively update the discovered reaction coordinate to enable the sampling of rare events with a feasible amount of computational time. That is, rounds of SPIB can be performed to iteratively optimize the discovered reaction coordinates …”

Employing Artificial Neural Networks to Identify Reaction Coordinates and Pathways for Self-Assembly

“…This bottleneck structure is a key attribute of the network design, which constrains the amount of information that can traverse the full network, forcing a learned compression of the input data. , As input data, we use the N BP = 15 processed BP distances ζ i and take the hyperparameters defining the weights of the loss terms as in ref . In contrast to TICA (and other approaches combining nonlinearity with time-lagged features ,, ), EncoderMap is an entirely structure-based approach.…”

“…57,58 As input data, we use the N BP = 15 processed BP distances ζ i and take the hyperparameters defining the weights of the loss terms as in ref 47. In contrast to TICA (and other approaches combining nonlinearity with time-lagged features 55,59,60 ), EncoderMap is an entirely structure-based approach.…”

AlphaFold2-RAVE: From sequence to Boltzmann ensemble

“…Choosing representative structures initialized from different clusters, we run multiple short unbiased molecular dynamics trajectories. We then use these trajectories as inputs for SPIB once again, this time using the past-future information bottleneck with a finite time lag 9,15 . Finally, we run biased metadynamics simulations using the information bottleneck learnt from this iteration of SPIB.…”