As COVID-19 became a global pandemic in 2020, a visualization of the entire structure of the novel virus became a fundamental concern to address. Through both advances in cryogenic electron microscopy and machine learning, the true backbone structures of proteins in viruses through learned patterns in alpha-carbon structures have been predicted in many cases. In this project, we address inaccuracies in alpha-carbon background structures in SARS-COV-2 protein structures using Graph Neural Network (GNNs) to solve background structures as edges in a graph of alpha-carbon nodes. We achieved high performance on a trained set of proteins. The code will be made available online under the project name.

The group would like to thank their faculty sponsor Jie Hou for their support with this project.