AI
Atelier: Learning Local Self-Supervised Features for CryoEM Volumes via Hypernetworks
Publication: cs.AI updates on arXiv.org
Author: Phillip Lo, Sudarshan Babu, Dari Kimanius, Aly A. Khan
Source published: 28/09/2026, 04:00:00 UTC
Briefing created: 28/09/2026, 04:12:29 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 28/09/2026, 04:03:25 UTC.
Briefings
The paper presents Atelier, a transformer-based hypernetwork that amortizes fitting implicit neural representations (INRs) for cryoEM maps. Pretrained on 5,439 Electron Microscopy Data Bank maps, Atelier generates high-fidelity reconstructions across diverse protein structures. Its key innovation is exposing a continuous, local feature field via intermediate activations, which serves as auxiliary channels for a 3D nested U-Net annotation head. This approach improves performance on eight voxel-level property prediction tasks compared to a volume-only baseline, demonstrating INRs as effective primitives for geometry-aware analysis.
Key points
- Atelier is a transformer-based hypernetwork that amortizes fitting implicit neural representations (INRs) for cryoEM maps.
- The framework was pretrained on 5,439 Electron Microscopy Data Bank maps to generate high-fidelity reconstructions across diverse protein structures.
Context and limitations
Results are limited to eight voxel-level property prediction tasks; the study does not report performance on other specific cryoEM metrics or compare against more extensive baselines.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 28/09/2026, 04:12:29 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifxhrah2qe3sr7uqxigrswey3qtzhpbkhoq4oj6qh7oguormntosq