Get started
Work

CryoAtlas: A Large Curated Dataset and Unified Benchmark for Cryo-EM Atomic Model Building

Year: 2026
Type: dataset
Abstract: CryoAtlas v1.0 is a curated dataset and unified benchmark for research on automated protein atomic model building from cryo-EM density maps. Derived from EMDB and PDB records available up to March 2026, it contains 17,809 curated map–model–sequence entries, comprising 16,657 training entries and 1,152 test entries. Following map–model alignment, sequence mapping, and quality filtering, each entry provides paired experimental density maps, atomic structures, and protein sequences.This ScienceDB Package brings together three complementary components:Training set: 16,657 entries distributed across four non-overlapping parts. Each part preserves complete EMD entries and includes density maps, atomic structures in CIF format, protein sequences in FASTA format, and metadata identifying its contents.Test set: 1,152 entries organized into two subsets: the 618-target benchmark used for comparative evaluation in the paper and the remaining 534 test entries. The 618 targets are a subset of the complete test set, not an additional dataset.Benchmark resources: Predicted atomic structures, per-target evaluation results, split definitions, and supporting documentation for reproducing and analyzing the benchmark. Evaluation covers structural accuracy, model-to-map agreement, and stereochemical quality.The package provides a unified entry point for downloading the training and test data and accessing the accompanying benchmark resources. The dataset is released under the Creative Commons Attribution–NonCommercial 4.0 International license (CC BY-NC 4.0); third-party materials retain their original licenses and attribution requirements. Additional documentation and code links are available at https://oulling.github.io/cryoatlas/.
Source: ScienceDB
DOI: 10.57760/sciencedb.0171m
Language: English
Open Access status: green
OpenAlex ID: W7224798654
Cites:
Cited by:
Related to: 10