Multiscale Enhanced Sampling Using Machine Learning

Abstract

Multiscale enhanced sampling (MSES) allows for an enhanced sampling of all-atom protein structures by coupling with the accelerated dynamics of the associated coarse-grained (CG) model. In this paper, we propose an MSES extension to replace the CG model with the dynamics on the reduced subspace generated by a machine learning approach, the variational autoencoder (VAE). The molecular dynamic (MD) trajectories of the ribose-binding protein (RBP) in both the closed and open forms were used as the input by extracting the inter-residue distances as the structural features in order to train the VAE model, allowing the encoded latent layer to characterize the difference in the structural dynamics of the closed and open forms. The interpolated data characterizing the RBP structural change in between the closed and open forms were thus efficiently generated in the low-dimensional latent space of the VAE, which was then decoded into the time-series data of the inter-residue distances and was useful for driving the structural sampling at an atomistic resolution via the MSES scheme. The free energy surfaces on the latent space demonstrated the refinement of the generated data that had a single basin into the simulated data containing two closed and open basins, thus illustrating the usefulness of the MD simulation together with the molecular mechanics force field in recovering the correct structural ensemble.

Enhancing Conformational Sampling for Intrinsically Disordered and Ordered Proteins by Variational Autoencoder

“…In this research, we constructed a VAE model, a generative neural network, which performed well on both IDPs and structured proteins. The VAE is trained to learn the sampled conformations from MD simulations in a relatively short period ( Table S4 ), which not only enables the reduction of the input protein structure with a nonlinear transformation towards low-dimensional space but also generates new conformations by disturbing the latent space with Gaussian noise [ 17 , 18 ]. A well-trained VAE can sample conformations of target proteins effectively and accurately, both for structured proteins and IDPs.…”

From Data to Knowledge: Systematic Review of Tools for Automatic Analysis of Molecular Dynamics Output

“…MDS in the context of MD was also described by Troyer and Cohen (1995) , Andrecut (2009) , Tribello and Gasparotto (2019) , and Srivastava et al (2020) . There are also examples of the application of other approaches: isometric feature mapping ( Stamati et al, 2010 ), kernel PCA ( Antoniou and Schwartz, 2011 ), diffusion map ( Rohrdanz et al, 2011 ; Zheng et al, 2011 ; Zheng et al, 2013a ; Zheng et al, 2013b ; Preto and Clementi, 2014 ), t-SNE ( Zhou et al, 2018 ; Zhou et al, 2019 ; Spiwok and Kříž, 2020 ), and VAE ( Hernández et al, 2018 ; Shamsi et al, 2018 ; Moritsugu, 2021 ; Tian et al, 2021 ).…”

Artificial intelligence guided conformational mining of intrinsically disordered proteins

“…Another recent breakthrough is the prediction of three-dimensional structures of proteins by neural network-based methods, Alphafold 4 and RoseTTafold 5 . With problems facing structured proteins being solved by these and other AI-based methods 6 – 9 , a new frontier is now presented by intrinsically disordered proteins (IDPs). Instead of adopting well-defined three-dimensional structures, IDPs readily access vast conformational space.…”

“…For structured proteins, autoencoders have been developed to represent structures in two-dimensional latent spaces and reconstruct the structures back in Cartesian coordinates 6 , 8 . In another recent study 9 , an autoencoder was trained to project the inter-residue distances of the ribose-binding protein into a two-dimensional latent space. The open and closed states of the protein were found to occupy separate regions in the latent space.…”