Artificial intelligence guided conformational mining of intrinsically disordered proteins

Abstract

Artificial intelligence recently achieved the breakthrough of predicting the three-dimensional structures of proteins. The next frontier is presented by intrinsically disordered proteins (IDPs), which, representing 30% to 50% of proteomes, readily access vast conformational space. Molecular dynamics (MD) simulations are promising in sampling IDP conformations, but only at extremely high computational cost. Here, we developed generative autoencoders that learn from short MD simulations and generate full conformational ensembles. An encoder represents IDP conformations as vectors in a reduced-dimensional latent space. The mean vector and covariance matrix of the training dataset are calculated to define a multivariate Gaussian distribution, from which vectors are sampled and fed to a decoder to generate new conformations. The ensembles of generated conformations cover those sampled by long MD simulations and are validated by small-angle X-ray scattering profile and NMR chemical shifts. This work illustrates the vast potential of artificial intelligence in conformational mining of IDPs.

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

“…By training the model with the first part, the model could learn enough diverse conformations, and a similar training process can be accomplished on a rather short MD trajectory. For the structure of AEs, we set the latent dimension to 2/3 ( : number of residues) and built them with four-layer encoders, referring to the setting designed by Gupta et al [ 12 ]. For VAEs, as the latent dimension was proved not to significantly affect the quality of generated conformations, we set the latent dimension to 2.…”

“…In this study, the scale of VAE-generated conformations was consistent with the test set size. As VAE applies uniform sampling to the latent space, when increasing the scale of generated conformations, we can foresee that the conformational features presented in the generated structures will be more continuous [ 12 ], and some high-energy conformations different from the input ones will be predicted. It is traditionally quite hard to sample these conformations with MD simulation, though it is becoming much easier for VAE.…”

“…By transforming the Cartesian coordinates of three-dimensional conformations into low-dimensional vectors, neural network models can be trained to generate the latent space from which new three-dimensional conformations can be sampled at a relatively low computational cost [ 11 ]. Gupta et al [ 12 ] designed an autoencoder (AE) for IDP trajectories, calculating a multivariate Gaussian distribution in latent space to sample more various IDP conformations. However, according to the Spearman correlation coefficient between the AE-generated conformations and the original ones, the Gaussian distribution might not conform to the true distribution of IDP conformations.…”

SPEADI: Accelerated Analysis of IDP-Ion Interactions from MD-Trajectories

“…This rate-limit should inform the choice of simulation techniques and parameters. Enhanced conformational sampling of IDPs beyond the sampling of REST2 simulations are being successfully developed in the form of variational autoencoders [ 104 ]. We are therefore pursuing the development of methods to speed up the equilibration of ions using machine learning techniques.…”

Initial Investigations of Intrinsically Disordered Regions in Inherited Retinal Diseases

“…Newer versions may show minor changes when compared with our findings. There have been studies assessing various deep-learning algorithms that predict disordered protein regions [ 23 , 24 ]. Eventually, our analysis may become automated as part of artificial intelligence efforts.…”