Structure modeling from density maps is a fundamental step for studying proteins and their complexes with cryogenic electron microscopy (cryo-EM). Although the resolution of cryo-EM density maps has improved in general, there are still many cases tracing protein main-chains in the map is challenging even in maps at near-atomic resolution. Here we developed a de novo protein structure modeling protocol, deepMainmast, which models protein structures for a cryo-EM density map that uses structure features detected by deep learning in the density map. Our proposed method also considers AlphaFold2 (AF2) models as external information. The main part of the DeepMainmast is a new de novo protein main-chain tracing method guided by predicted positions of Ca atoms and types of amino acids by deep learning. Compared with its predecessor, MAINMAST, the improvement includes the use of deep learning and a different, more effective main-chain tracing approach called the Vehicle Routing Problem solver and Constraint Problem Solver. On top of structure models generated by deepMainmast, we run AF2 and blend AF2 models with models from the deepMainmast if they have local consistency with the density map. The protocol generates up to twenty models in total, with different levels of blending of models. These are finally prioritized by the DAQ score, a model quality assessment score we have recently developed. deepMainmast was shown to outperform existing methods on a benchmark dataset substantially. When compared with AF2, deepMainmast alone showed higher accuracy on more maps among the 178 high-resolution maps in the dataset. Moreover, the entire protocol substantially outperforms AF2.

deepMainmast: <i>De novo</i> protein structure modeling for cryo-EM with deep learning and structure prediction

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

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

deepMainmast: De novo protein structure modeling for cryo-EM with deep learning and structure prediction