CryoID2: deep learning-enhanced identification of protein and RNA from cryoEM maps

CryoID2 identifies molecular components in cryoEM/cryoET maps by predicting structure features and ranking sequence candidates, achieving near-reference fitting quality on a 52-case benchmark.

• Why it matters: Molecular identification of endogenous or poorly characterized cryoEM/cryoET complexes is a bottleneck, especially when sequence information is limited.
• What they did: CryoID2 uses three 3D U-Nets to predict structure classes, key-atom positions, and reduced residue/nucleotide types, then traces backbone and generates short sequence queries ranked with BLASTp for proteins and a maximin strategy for RNA. It retrieves matching PDB or AlphaFold models, rigidly fits them using predicted features and density, and ranks by map agreement.
• The result: Module tests show high-recall feature prediction and resolution-dependent atom clustering, and the 52-case benchmark recovers near-reference fitting quality; it correctly identified PfA-M17 from a 166-protein mass-spectrometry pool and unannotated Ca-raiA RNA from a genome-derived non-coding sequence set.

AI summary of the abstract · may contain errors · not medical advice · read the paper.