Atelier creates detailed local features for cryoEM protein maps
Atelier: Learning Local Self-Supervised Features for CryoEM Volumes via Hypernetworks
Artificial Intelligence
Summary
Cryo-electron microscopy (cryoEM) produces 3D maps of proteins that need detailed features to understand their structure. The authors developed Atelier, a system that learns to represent these maps as smooth functions rather than fixed grids, making it easier to analyze proteins at different scales. Atelier uses a special neural network trained on thousands of maps to quickly generate consistent features across proteins, improving the accuracy of identifying detailed properties in the maps. This approach helps researchers interpret complex protein structures more effectively.
What this means in practice
- •For structural biology teams: Produce fine-grained, continuous representations of cryoEM maps to improve protein structure annotation accuracy across diverse samples.
- •For medical imaging engineers: Adapt the method for generating scale-independent local features to analyze 3D medical imaging data with complex spatial structures.
Authors
Phillip Lo, Sudarshan Babu, Dari Kimanius, Aly A. Khan
Abstract
CryoEM map interpretation requires features that are spatially localized, consistent across samples, and informative across spatial scales. Most deep learning methods for map annotation extract features from fixed voxel grids. However, implicit neural representations (INRs) are able to model volumetric data as scale-agnostic, coordinate-conditioned functions. INRs are therefore attractive for cryoEM, but fitting a separate INR for each map is too expensive for large-scale feature extraction and produces representations that are not aligned across samples. We introduce Atelier, a self-supervised framework that amortizes INR fitting for reconstructed cryoEM maps. Pretrained on 5,439 Electron Microscopy Data Bank maps, Atelier is a transformer-based hypernetwork that generates high-fidelity reconstructions across a wide range of protein structures, including large multi-subunit assemblies. Beyond reconstruction, the INR generated by the pretrained transformer exposes a continuous, local feature field through its intermediate activations at any spatial query point, a property that voxel grid and patch-tokenizer architectures do not naturally provide. Used as auxiliary channels to a 3D nested U-Net annotation head trained from scratch, these coordinate-conditioned features improve performance on eight voxel-level property prediction tasks over a volume-only baseline. Our results demonstrate that amortized implicit neural representations are an effective primitive for geometry-aware analysis of cryoEM data.