Learning protein conformational space by enforcing physics with convolutions and latent interpolations

TL;DR

This paper introduces a convolutional neural network that learns a continuous, physically plausible protein conformational space, enabling accurate prediction of transition paths and transfer learning across proteins with minimal data.

Contribution

The authors develop a novel CNN-based method that models protein conformational spaces continuously and enforces physical plausibility, improving transition path prediction and transferability.

Findings

Successfully predicts biologically relevant transition paths without explicit path examples.

Enables transfer learning of features between different proteins.

Achieves high performance with limited training data.

Abstract

Determining the different conformational states of a protein and the transition paths between them is key to fully understanding the relationship between biomolecular structure and function. This can be accomplished by sampling protein conformational space with molecular simulation methodologies. Despite advances in computing hardware and sampling techniques, simulations always yield a discretized representation of this space, with transition states undersampled proportionally to their associated energy barrier. We present a convolutional neural network that learns a continuous conformational space representation from example structures, and loss functions that ensure intermediates between examples are physically plausible. We show that this network, trained with simulations of distinct protein states, can correctly predict a biologically relevant non-linear transition path, without any example on the path provided. We also show we can transfer features learnt from one protein to others, which results in superior performances, and requires a surprisingly small number of training examples.