Coupling Molecular Dynamics and Deep Learning to Mine Protein Conformational Space

Matteo T Degiacomi

doi:10.1016/j.str.2019.03.018

Abstract

Flexibility is often a key determinant of protein function. To elucidate the link between their molecular structure and role in an organism, computational techniques such as molecular dynamics can be leveraged to characterize their conformational space. Extensive sampling is, however, required to obtain reliable results, useful to rationalize experimental data or predict outcomes before experiments are carried out. We demonstrate that a generative neural network trained on protein structures produced by molecular simulation can be used to obtain new, plausible conformations complementing pre-existing ones. To demonstrate this, we show that a trained neural network can be exploited in a protein-protein docking scenario to account for broad hinge motions taking place upon binding. Overall, this work shows that neural networks can be used as an exploratory tool for the study of molecular conformational space.

Highlights

Function at the molecular level emerges from the arrangement of individual atoms and their associated dynamics
We demonstrate that autoencoders can generate new, realistic protein conformations complementing pre-existing data produced by molecular dynamics (MD) simulations
To generate low-dimensional representations of proteins’ conformational space, we exploit an autoencoder (Figure 1). This is a type of neural network that attempts to first compress and decompress a multidimensional input, so that the difference between input and output is minimized

Summary

Introduction

Function at the molecular level emerges from the arrangement of individual atoms and their associated dynamics. Atomic information can be obtained via nuclear magnetic resonance This technique informs about dynamics and can sometimes identify multiple states, given that the molecule of interest is not too large. Other techniques report near-atomic or lower-resolution data, describing for instance the shape of the protein (e.g., electron microscopy, ion mobility-mass spectrometry, or small-angle X-ray scattering), or measuring specific interatomic distances (e.g., chemical crosslinking). Even considering this broad palette of techniques, studying the structure of a protein featuring multiple states is usually challenging, and even when a single conformation is present its thermal fluctuations may render data interpretation arduous

Results

Discussion

Conclusion