MICA 2026 benchmark dataset: 30 novel cryo-EM maps, AlphaFold3 inputs and evaluation outputs
Year: 2026
Type: dataset
Abstract: Test dataset and raw evaluation output for a benchmark of MICA (Gyawali, Dhakal & Cheng, Communications Chemistry 2025) against EModelX(+AF), ModelAngelo, CryoAtom and EMProt for atomic protein model building from cryo-EM maps. The test set contains 30 single particle cryo-EM maps (1.95 to 3.92 Å) released in EMDB/PDB in 2026. Every protein chain of at least 30 residues has no more than 25% sequence identity to any PDB entry released before 2026, so none of the benchmarked methods can have been trained on these proteins or close homologs. Entries were deduplicated at 25% identity and sampled with a fixed seed (2026), 10 per size bin. Files: dataset_entries.zip (maps, sequences, deposited reference structures, biological assemblies), af3_predictions.zip (AlphaFold Server predictions used as inputs; subject to the AlphaFold Server Output Terms of Use, non-commercial use only), scoring_raw_outputs.zip (raw US-align and Phenix chain_comparison output for every model), code_and_tables.zip (snapshot of the code repository with selection tables, scores and protocol).
Author: Dodia, Panam
Language: English
Subfield: Structural Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7223375402