Molecular insights from conformational ensembles via machine learning

Abstract

Biomolecular simulations are intrinsically high dimensional and generate noisy datasets of everincreasing size. Extracting important features in the data is crucial for understanding the biophysical properties of molecular processes, but remains a big challenge. Machine learning (ML) provides powerful dimensionality reduction tools. However, such methods are often criticized to resemble black boxes with limited human-interpretable insight.We use methods from supervised and unsupervised ML to efficiently create interpretable maps of important features from molecular simulations. We benchmark the performance of several methods including neural networks, random forests and principal component analysis, using a toy model with properties reminiscent of macromolecular behavior. We then analyze three diverse biological processes: conformational changes within the soluble protein calmodulin, ligand binding to a G protein-coupled receptor and activation of an ion channel voltage-sensor domain, unravelling features critical for signal transduction, ligand binding and voltage sensing. This work demonstrates the usefulness of ML in understanding biomolecular states and demystifying complex simulations. 1

Critical interactions for SARS-CoV-2 spike protein binding to ACE2 identified by machine learning

“…A previous study using MLP to identify important residues between different protein configurations showed that more hidden layers led to lower accuracy. 34 Here, we reached an accuracy of 1 for classifying SARS-CoV from SARS-CoV-2 with only one hidden layer in the MLP implementation. If we remove all hidden layers of an MLP model, it will have the same architecture as a linear model, such as the LR approach also used here, for which the accuracy was 1 as well.…”

Recognition of a Tandem Lesion by DNA Glycosylases Explored Combining Molecular Dynamics and Machine Learning

“…Their power in finding important information out of large amount of data has been exploited by the biochemistry community, many interesting applications have been showcased in the literature. Recently, Fleetwood et al 30 have demonstrated its capability in learning ensemble properties from molecular simulations and providing easily interpretable metrics describing important structural or chemical features. The machine-learning analysis of our trajectories is based on the demystifying package from Fleetwood et al 30 .…”

“…Recently, Fleetwood et al 30 have demonstrated its capability in learning ensemble properties from molecular simulations and providing easily interpretable metrics describing important structural or chemical features. The machine-learning analysis of our trajectories is based on the demystifying package from Fleetwood et al 30 . Residues highlighted as providing a significant contribution to the MutM:DNA bonding by the MLP analysis are in agreement with data from the literature.…”

“…The rectified linear unit function (ReLU) 54 was used for the activation of neurons, and the Adam algorithm was used for optimization. The inverse of the distances between the geometric centers of the residues were used as the input features for the multilayer perceptrons neural network, due to better overall performance over Cartesian coordinates, according to Fleetwood et al 30 . These internal coordinates were computed for all residue pairs and all frames.…”

“…This first local analysis suggests that the singly-vs. tandem-damaged 13-bp duplex present different interaction patterns, with non trivial changes in the binding mode and its dynamics. In order to probe more extensively the structural and dynamic consequences of dG20 → Ap20 substitution, we have relied on a recently-proposed machine-learning protocol 30 to identify other residues possibly implied in the recognition mechanism.…”