Research map / Probabilistic and scientific modeling / Variational and generative sampling
Empirical Bayes Flow Matching for Continuous Cryo-EM Heterogeneity
Daniel Aibinder, Azmi Haider, Dan Rosenbaum
Read the paperCheck your idea against 137,217 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 Empirical Bayes Flow Matching to learn generative priors over latent atomic structures from indirect, noisy cryo-EM projection images for continuous heterogeneity.
Problem
Learning generative priors over high-dimensional latent 3D atomic structures from noisy 2D cryo-EM projection images generated by a non-linear forward operator, where the structures are never directly observed.
Approach
Combines empirical Bayes with flow matching so the generative prior over latent atomic structures is learned directly from indirect, noisy projection observations rather than from clean structures; the abstract is truncated so the precise mechanism is not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add persistent latents updated by stochastic approximation (ablation) | EB-FM · Two-moons, measurement dim M=8 | manifold recovery | Improved |
| Add persistent latents (ablation) | EB-FM · MNIST (50 measurements) | EMD PSNR: 13.75 -> 15.67 | Improved |
| Use flow matching with gradient guidance instead of diffusion CG solver (DiEM-iso) | EB-FM · MNIST (50 measurements) | EMD PSNR: 13.75 -> 15.67 | Improved |
| Use gradient-guided EB-FM instead of DiEM-iso on two-moons | EB-FM · Two-moons under-determined regime | mode preservation | Improved |
| Remove backbone continuity loss (ablation) | EB-FM · Cryo-EM 1a62_A | EMD-RMSD (A): 1.62 -> 3.00 | Hurt |
| Remove backbone continuity loss (ablation) | EB-FM · Cryo-EM 1egw_C | EMD-RMSD (A): 6.22 -> 9.12 | Hurt |
| Randomize initialization of persistent latents (ablation) | EB-FM · Cryo-EM (3 of 4 proteins) | distributional metrics: 1egw_C EMD-RMSD 6.22 -> 5.56 | Improved |
| Randomize initialization of persistent latents (ablation) | EB-FM · Cryo-EM | pairwise RMSD / RMSF flexibility magnitude: 1a62_A pairwise RMSD 1.58 -> 1.43 | No clear effect |
| Add small pose/CTF error to all particles (ablation) | EB-FM · Cryo-EM 1egw_C | EMD-RMSD (A): 6.22 -> 4.73 | Improved |
| Add pose/CTF error to 30% of particles (ablation) | EB-FM · Cryo-EM 1a62_A | EMD-RMSD (A): 1.62 -> 2.06 | Hurt |
| Train prior jointly with latents instead of direct latent optimization alone (ablation) | EB-FM · Cryo-EM latent RMSD | latent RMSD stability | Improved |
| Use EB-FM prior vs cryoDRGN-guided-AF3 on EMD-RMSD | EB-FM · Cryo-EM 4 proteins | EMD-RMSD: 1a62_A 1.80 -> 1.62 | Improved |
Experimental setup
Builds on: flow matching
Closest papers by meaning
- Reconstructing Heterogeneous Biomolecules via Hierarchical Gaussian Mixtures and Part Discovery (NeurIPS 2025 · similarity 0.89)
- CryoFM: A Flow-based Foundation Model for Cryo-EM Densities (ICLR 2025 · similarity 0.89)
- Mixture of neural fields for heterogeneous reconstruction in cryo-EM (NeurIPS 2024 · similarity 0.88)
- Multiscale guidance of protein structure prediction with heterogeneous cryo-EM data (NeurIPS 2025 · similarity 0.88)
- CryoBench: Diverse and challenging datasets for the heterogeneity problem in cryo-EM (NeurIPS 2024 · similarity 0.88)
- cryoSPHERE: Single-Particle HEterogeneous REconstruction from cryo EM (ICLR 2025 · similarity 0.87)
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.
