Iterative model-based density improvement yields better atomic structures from cryo-EM maps

Atomic models based on high-resolution density maps are the ultimate result of the cryo-EM structure determination process. Current cryo-EM model refinement procedures work with the experimental density map that remains constant throughout the process. Here, we introduce a general procedure that iteratively improves cryo-EM density maps based on prior knowledge of an atomic reference structure. The procedure optimizes contrast of cryo-EM densities by local amplitude scaling (LocScale) based on an atomic model without introducing model bias. We alternate the procedure with consecutive rounds of model refinement and tested it on four cryo-EM structures of TRPV1, β-galactosidase, γ-secretase and RNA polymerase III. We demonstrate that LocScale density improvement reveals previously undiscovered map features and improves the quality of atomic models. The presented approach enhances the interpretability of cryo-EM density maps and provides an implementation reminiscent of iterative density improvement as it is routinely employed in the refinement of X-ray crystallographic models.