Graph based electron density guided atomic recovery

Abstract

Here we propose Graph based ELectron density guided ATOmic recovery model (GELATO) which takes in a processed 3D electron density map and aims to provide the complete structure of the constituent protein structure. We show that we can detect the positions of residues with much higher recall than before while maintaining comparable precision. We show that we are on par with the state-of-the-art method, CryoAtom, for amino acid identiőcation, while also being 5 times lighter (in terms of the number of parameters) and an order of magnitude faster than CryoAtom in the time required to process the input. Lastly we show our preliminary works on the structure determination models and show that we achieve competitive backbone recovery.