CryoSPIN: Improving Ab-Initio Cryo-EM Reconstruction with Semi-Amortized Pose Inference

Shayan Shekarforoush,David B. Lindell,Marcus A. Brubaker,David J. Fleet

2024-10-03

Abstract:Cryo-EM is an increasingly popular method for determining the atomic resolution 3D structure of macromolecular complexes (eg, proteins) from noisy 2D images captured by an electron microscope. The computational task is to reconstruct the 3D density of the particle, along with 3D pose of the particle in each 2D image, for which the posterior pose distribution is highly multi-modal. Recent developments in cryo-EM have focused on deep learning for which amortized inference has been used to predict pose. Here, we address key problems with this approach, and propose a new semi-amortized method, cryoSPIN, in which reconstruction begins with amortized inference and then switches to a form of auto-decoding to refine poses locally using stochastic gradient descent. Through evaluation on synthetic datasets, we demonstrate that cryoSPIN is able to handle multi-modal pose distributions during the amortized inference stage, while the later, more flexible stage of direct pose optimization yields faster and more accurate convergence of poses compared to baselines. On experimental data, we show that cryoSPIN outperforms the state-of-the-art cryoAI in speed and reconstruction quality.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper attempts to address several key issues in single-particle 3D structure reconstruction in cryo-electron microscopy (cryo-EM). Specifically: 1. **Multimodal Pose Distribution Problem**: In the initial stage, since the structure is not yet fully determined, the pose distribution of particles is often multimodal. Existing deep learning-based methods typically use amortized inference to predict poses, but this approach may not accurately represent the multimodal posterior pose distribution. 2. **Pose Optimization Accuracy Problem**: As the reconstruction process progresses, the pose distribution gradually becomes unimodal, requiring more precise pose optimization. However, amortized inference methods may get trapped in local optima, leading to inaccurate pose estimation. 3. **Computational Efficiency Problem**: Existing methods have low computational efficiency when handling large-scale datasets, especially during the pose optimization stage. To address these issues, the paper proposes a new semi-amortized method called cryoSPIN. This method combines the advantages of amortized inference and direct pose optimization, allowing it to handle multimodal pose distributions in the initial stage and improve pose estimation accuracy through direct optimization in the later stages, thereby achieving faster and higher-quality 3D structure reconstruction.

CryoSPIN: Improving Ab-Initio Cryo-EM Reconstruction with Semi-Amortized Pose Inference

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

Equivariant amortized inference of poses for cryo-EM

Amortized Inference for Heterogeneous Reconstruction in Cryo-EM

Improved Cryo-EM Pose Estimation and 3D Classification through Latent-Space Disentanglement

Amortized template-matching of molecular conformations from cryo-electron microscopy images using simulation-based inference

cryoSPHERE: Single-particle heterogeneous reconstruction from cryo EM

CryoChains: Heterogeneous Reconstruction of Molecular Assembly of Semi-flexible Chains from Cryo-EM Images

CryoFormer: Continuous Heterogeneous Cryo-EM Reconstruction using Transformer-based Neural Representations

Towards Interpretable Cryo-EM: Disentangling Latent Spaces of Molecular Conformations

Overcoming the preferred orientation problem in cryoEM with self-supervised deep-learning

End-to-end neural ab initio reconstruction of challenging cryo-EM and cryo-ET datasets

Boosted ab initio Cryo-EM 3D Reconstruction with ACE-EM

Overcoming the preferred-orientation problem in cryo-EM with self-supervised deep learning

Cryo-ZSSR: multiple-image super-resolution based on deep internal learning

Efficient high-resolution refinement in cryo-EM with stochastic gradient descent

Robust ab initio solution of the cryo-EM reconstruction problem at low resolution with small data sets

Multiple-image super-resolution of cryo-electron micrographs based on deep internal learning

End-to-End Simultaneous Learning of Single-particle Orientation and 3D Map Reconstruction from Cryo-electron Microscopy Data

A strategy combining denoising and cryo-EM single particle analysis

Novel Artificial Intelligence-Based Approaches for Ab Initio Structure Determination and Atomic Model Building for Cryo-Electron Microscopy