Abstract:Cryo-electron microscopy (cryo-EM) is a powerful technique in structural biology and drug discovery, enabling the study of biomolecules at high resolution. Significant advancements by structural biologists using cryo-EM have led to the production of over 38,626 protein density maps at various resolutions1. However, cryo-EM data processing algorithms have yet to fully benefit from our knowledge of biomolecular density maps, with only a few recent models being data-driven but limited to specific tasks. In this study, we present CryoFM, a foundation model designed as a generative model, learning the distribution of high-quality density maps and generalizing effectively to downstream tasks. Built on flow matching, CryoFM is trained to accurately capture the prior distribution of biomolecular density maps. Furthermore, we introduce a flow posterior sampling method that leverages CRYOFM as a flexible prior for several downstream tasks in cryo-EM and cryo-electron tomography (cryo-ET) without the need for fine-tuning, achieving state-of-the-art performance on most tasks and demonstrating its potential as a foundational model for broader applications in these fields.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to address several key issues in cryo-electron microscopy (cryo-EM) data processing: 1. **Generation of High-Quality Density Maps**: Existing cryo-EM data processing algorithms have not fully utilized the knowledge of biomolecular density maps, and most models are limited to specific tasks. This paper proposes a flow-matching-based generative model (CRYOFM) that can learn the distribution of high-quality density maps and perform well in downstream tasks. 2. **Flexibility in Downstream Tasks**: Existing data-driven models are usually designed for specific tasks, limiting their generality and flexibility. CRYOFM introduces a flow posterior sampling method, allowing it to be applied to multiple downstream tasks such as noise removal, anisotropic noise denoising, and missing wedge recovery without fine-tuning. 3. **Modeling Prior Distributions**: In cryo-EM data processing, prior distributions are crucial for guiding the reconstruction process and improving structural accuracy. CRYOFM provides a more powerful and expressive data-driven prior by learning the distribution of high-resolution density maps. 4. **Utilization of Experimental Electron Density Maps**: Although foundational models have made significant progress in protein structure prediction and design, the application of experimental electron density maps in this field is relatively limited. CRYOFM fills this gap, demonstrating its potential in cryo-EM and cryo-electron tomography (cryo-ET). ### Main Contributions 1. **Proposing the First Flow-Based Generative Model**: CRYOFM is the first foundational model to learn the distribution of high-quality cryo-EM density maps. 2. **Introducing a Flow Posterior Sampling Algorithm**: This algorithm enables CRYOFM to serve as a flexible prior, applicable to various downstream tasks. 3. **Outstanding Performance in Multiple Downstream Tasks**: Without fine-tuning, CRYOFM achieves state-of-the-art results in several experiments, particularly in noise removal and missing wedge recovery tasks. 4. **Exploring Different Model Architectures and Configurations**: The generative model is optimized for training on large-scale biomolecular density maps. ### Experimental Validation The paper validates the effectiveness of CRYOFM through the following experiments: 1. **Spectral Noise Denoising**: CRYOFM shows higher robustness and lower failure rates at different noise levels, significantly improving the quality and resolution of density maps. 2. **Anisotropic Noise Denoising**: CRYOFM outperforms baseline models in anisotropic noise denoising tasks, successfully recovering signals without introducing significant bias. 3. **Missing Wedge Recovery**: By simulating the missing wedge effect, CRYOFM effectively recovers the original signal, reducing missing wedge artifacts. 4. **De Novo Modeling**: In the task of reconstructing coarse density maps from 2D particle projections, CRYOFM demonstrates performance comparable to or better than existing methods. In summary, by learning the distribution of high-quality cryo-EM density maps, CRYOFM provides a powerful foundational model that excels in multiple downstream tasks, offering new tools and methods for the development of the cryo-EM and cryo-ET fields.

CryoFM: A Flow-based Foundation Model for Cryo-EM Densities

A comprehensive foundation model for cryo-EM image processing

Protein complex structure modeling by cross-modal alignment between cryo-EM maps and protein sequences

A New Protocol for Atomic-Level Protein Structure Modeling and Refinement Using Low-to-Medium Resolution Cryo-EM Density Maps

CryoRes: Local Resolution Estimation of Cryo-EM Density Maps by Deep Learning.

Enhancing Density Maps by Removing the Majority of Particles in Single Particle Cryogenic Electron Microscopy Final Stacks

CryoAI: Amortized Inference of Poses for Ab Initio Reconstruction of 3D Molecular Volumes from Real Cryo-EM Images

DiffModeler: Large Macromolecular Structure Modeling in Low-Resolution Cryo-EM Maps Using Diffusion Model

CryoSTAR: leveraging structural priors and constraints for cryo-EM heterogeneous reconstruction

De Novo Atomic Protein Structure Modeling for Cryo-EM Density Maps Using 3D Transformer and Hidden Markov Model

Cryo-EM structure and B-factor refinement with ensemble representation

Cryo2StructData: A Large Labeled Cryo-EM Density Map Dataset for AI-based Modeling of Protein Structures

Latent Space Diffusion Models of Cryo-EM Structures

Improvement of cryo-EM maps by density modification

Computational models of amorphous ice for accurate simulation of cryo-EM images of biological samples

Improved automated model building for cryo-EM maps using CryFold

CryoJAM: Automating Protein Homolog Fitting in Medium Resolution Cryo-EM Density Maps

DiffModeler: large macromolecular structure modeling for cryo-EM maps using a diffusion model

Atomic Accuracy Models from 4.5 Å Cryo-Electron Microscopy Data with Density-Guided Iterative Local Refinement

CryoAlign: feature-based method for global and local 3D alignment of EM density maps