De novo model building with cryo-EM density and coevolution restraints

Abstract

Low-resolution limitations have long plagued cryo-EM de novo modeling. To address this challenge, I modified Rosetta de novo to incorporate fragments with tertiary context, or nonlocal fragments. Limitations I identified led me to develop a novel method for low-resolution structure determination, combining density and predicted restraints. I used my pipeline to solve and publish two difficult structures, and show it can handle modeling homologous conformers better than attention-based deep-learning methods, AlphaFold and RosettaFold.