Despite recent advancements in the development of deep learning methods for protein structure prediction and representation, few works have focused on the simultaneous utilization and prediction of protein sidechain and protein backbone structure information. We present SidechainNet, a dataset and collection of machine learning software tools that directly builds on the previously developed ProteinNet dataset, but includes torsional angle and atomic coordinate information capable of completely describing all heavy atoms of each protein structure. We argue for the inclusion of sidechain information, describe the process by which we organize it, and provide a software library for its acquisition, handling, and subsequent training with machine learning models. Lastly, we present our work on manipulating all-atom structures from SidechainNet directly with molecular dynamics software for energy-based minimization and machine learning model training.

SidechainNet: An all-atom protein structure dataset for machine learning

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.