Research map / Probabilistic and scientific modeling / Molecules and proteins
cryoSPHERE: Single-Particle HEterogeneous REconstruction from cryo EM
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes cryoSPHERE, a VAE that learns a Gaussian-mixture segmentation of an AlphaFold base structure and per-image rigid-body segment motions to reconstruct conformational ensembles from noisy cryo-EM images.
Problem
Heterogeneous cryo-EM reconstruction recovers only consensus or volume-level conformations, and structure-based methods get stuck in local minima under the high noise and 2D-projection ambiguity of low-SNR data.
Approach
Learn, end-to-end, a soft segmentation of the amino-acid chain (via a temperature-annealed Gaussian mixture over residue indices) and move the segments as approximately rigid bodies per image, with a VAE encoder mapping each image to a latent that decodes to segment rotations and translations.
Key results
- On the MD phytochrome dataset at SNR 0.001, cryoSPHERE reaches a predicted-vs-true domain distance correlation of 0.73-0.74 and outperforms cryoDRGN and cryoStar on mean FSC at the 0.5 and 0.143 cutoffs.
- At SNR 0.1 cryoStar slightly outperforms cryoSPHERE on FSC, while at SNR 0.01 cryoSPHERE outperforms cryoStar.
- On EMPIAR-12093 Pfr, cryoStar produces unphysical, clashing motions whereas cryoSPHERE recovers structured scissor-like opening without clashes.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use cryoSPHERE instead of cryoDRGN | Molecular dynamics phytochrome reconstruction · Molecular dynamics phytochrome, SNR 0.001 | FSC at the 0.143 cutoff | Helped |
| Use cryoSPHERE instead of cryoStar | Molecular dynamics phytochrome reconstruction · Molecular dynamics phytochrome, SNR 0.001 | FSC at the 0.143 cutoff | Helped |
| Use cryoSPHERE instead of cryoDRGN | Molecular dynamics phytochrome reconstruction · Molecular dynamics phytochrome, SNR 0.1 | FSC | Helped |
| Use cryoSPHERE instead of cryoStar | Molecular dynamics phytochrome reconstruction · Molecular dynamics phytochrome, SNR 0.1 | FSC | Hurt |
| Use cryoSPHERE instead of cryoDRGN | Molecular dynamics phytochrome reconstruction · Molecular dynamics phytochrome, SNR 0.01 | FSC | Helped |
| Use cryoSPHERE instead of cryoStar | Molecular dynamics phytochrome reconstruction · Molecular dynamics phytochrome, SNR 0.01 | FSC | No clear effect |
| Use cryoSPHERE instead of cryoStar | Phytochrome Pfr reconstruction · EMPIAR-12093 Pfr | physical plausibility of reconstructed conformations | Helped |
| Use cryoSPHERE instead of cryoStar | Phytochrome Pfr reconstruction · EMPIAR-12093 Pfr | recovery of scissoring motion | Helped |
| Increase the number of segments (ablation) | cryoSPHERE · Molecular dynamics phytochrome, SNR 0.1, 0.01, and 0.001 | FSC | Helped |
| Increase the number of segments (ablation) | cryoSPHERE · Molecular dynamics phytochrome, lower SNR | FSC | Helped |
| Move protein segments as rigid bodies rather than moving each residue individually | cryoSPHERE compared with cryoStar · Molecular dynamics phytochrome, SNR 0.1, 0.01, and 0.001 | FSC | Mixed |
| Use the cryoStar volume method instead of cryoDRGN | Volume reconstruction using the structural method's latent space · Molecular dynamics phytochrome, SNR 0.1, 0.01, and 0.001 | FSC | No clear effect |
Surprising findings
- Volume-based debiasing fails for the Pfr phytochrome state because the averaged density signal is too weak to recover the upper domains.
- At high SNR (0.1), the per-residue cryoStar method slightly beats the segment-based cryoSPHERE on FSC.
Limitations
- Relies on an accurate nominal structure (e.g., AlphaFold) and a consensus backprojection fit; the base structure can bias results, requiring a volume-based debiasing step.
- Volume-based debiasing fails for the most dynamic and noisy state (Pfr), so conformations cannot always be validated.
- Assumes poses are known from off-the-shelf estimation, and cannot remove extra residues present in the base structure by design.
Experimental setup
Datasets: Synthetic toy phytochrome dataset (AlphaFold-multimer structure, UniProt Q9RZA4) Molecular dynamics phytochrome dataset (PDB 4Q0J, metadynamics simulations) EMPIAR-10180 EMPIAR-12093 PDB 5NRL
Benchmarks: Synthetic phytochrome MD dataset (SNR 0.1, 0.01, 0.001) EMPIAR-10180 EMPIAR-12093
Metrics: Fourier shell correlation at 0.5 and 0.143 cutoffs predicted vs true inter-domain distance correlation clash counts
Baselines: cryoDRGN cryoStar cryoSPARC RELION 3DFlex
Builds on: cryoDRGN cryoStar AlphaFold DRGN-AI VAE
Closest papers by meaning
- CryoBench: Diverse and challenging datasets for the heterogeneity problem in cryo-EM (NeurIPS 2024 · similarity 0.93)
- Multiscale guidance of protein structure prediction with heterogeneous cryo-EM data (NeurIPS 2025 · similarity 0.92)
- Mixture of neural fields for heterogeneous reconstruction in cryo-EM (NeurIPS 2024 · similarity 0.92)
- Reconstructing Heterogeneous Biomolecules via Hierarchical Gaussian Mixtures and Part Discovery (NeurIPS 2025 · similarity 0.91)
- CryoSPIN: Improving Ab-Initio Cryo-EM Reconstruction with Semi-Amortized Pose Inference (NeurIPS 2024 · similarity 0.91)
- CryoFM: A Flow-based Foundation Model for Cryo-EM Densities (ICLR 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.