CryoACE

Overview

Atom-centric reconstruction for static and dynamic cryo-EM structures.

Protein automodeling from cryo-EM density maps faces unique challenges in enforcing physicochemical validity and managing conformational heterogeneity. CryoACE reconstructs precise atomic graphs for both homogeneous and heterogeneous structures with an atom-centric reconstruction paradigm, where density features are sampled directly at atomic coordinates and recycled iteratively to refine structures.

CryoACE replaces expensive voxel convolutions with efficient multimodal fusion and uses training-free guidance driven by predicted local resolution priors to resolve dynamic ambiguity. On a newly constructed high-quality dataset, CryoACE outperforms existing static baselines and reveals atomic-level dynamic conformations on real-world datasets including EMPIAR-10345 without relying on pre-built static structures.

10,915 curated density-structure-sequence triplets

Static + dynamic homogeneous and heterogeneous model building

Atom-level coordinate generation directly conditioned on density maps

Method

Density features are sampled at predicted atoms, then recycled for refinement.

CryoACE atom-centric model building pipeline. — CryoACE combines sequence priors, atom-centric density sampling, iterative recycling, auxiliary quality prediction, and training-free guidance for controllable model building.

Results

Improved static reconstruction and heterogeneous ensemble recovery.

Quantitative comparison for homogeneous model building. — Homogeneous model building benchmark across geometric accuracy, completeness, and map-model fit.

Heterogeneous model building results on alpha V beta 8 Integrin. — Heterogeneous modeling on alpha V beta 8 Integrin, showing atomic-level dynamic conformations.

Citation

BibTeX

@inproceedings{li2026cryoace,
  title = {CryoACE: An Atom-centric Framework for Accurate and Automated Model Building in Cryo-EM},
  author = {Li, Minzhang and Li, Mingrui and Qin, Weichen and Chen, Qihe and Shen, Sixian and Pei, Yuan and Zhang, Jiakai and Yu, Jingyi},
  booktitle = {Proceedings of the International Conference on Machine Learning},
  year = {2026}
}