Cryo-EM can now image molecular machines at near-atomic detail — but turning a blurry 3D density map into an atomic model used to mean months of expert handwork. Today’s digest is about the deep learning that automates that journey, step by step. Bepler et al.’s Topaz finds particles in noisy micrographs, work that otherwise ‘can take months of manual effort.’ Sanchez-Garcia et al.’s DeepEMhancer sharpens maps in ‘a single step.’ Maddhuri Venkata Subramaniya et al.’s Emap2sec reads secondary structure at ‘5 to 10 Å.’ Pfab et al.’s DeepTracer does ‘fast de novo’ backbone modeling. Jamali et al.’s ModelAngelo builds atomic models ‘of similar quality to those generated by human experts’ — and beats them at identifying unknown proteins. And Giri & Kihara’s CryoREAD extends it to DNA and RNA. From blob to blueprint, automatically.
