Learning protein conformational space by enforcing physics with convolutions and latent interpolations
This addresses the challenge of efficiently modeling protein dynamics for structural biology, offering a method to infer transition states from limited simulation data, though it is incremental as it builds on existing neural network approaches.
The authors tackled the problem of undersampling protein conformational transitions in molecular simulations by developing a convolutional neural network that learns a continuous representation from example structures, enforcing physical plausibility in intermediates. They showed it can predict a biologically relevant non-linear transition path without examples on that path and transfer features between proteins with few training examples.
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.