GEM: 3D Gaussian Splatting for Efficient and Accurate Cryo-EM Reconstruction

2025-09-29

AI Summary

Cryo-electron microscopy (cryo-EM) 3D reconstruction faces prohibitive computational and memory costs due to large-scale particle image datasets: Fourier-based methods suffer from limited fidelity, while real-space approaches—such as neural radiance fields (NeRFs)—exhibit cubic complexity in volume resolution. To address this, we propose a novel real-space density modeling paradigm based on 3D Gaussian lattices, replacing conventional voxels or NeRFs with compact, differentiable Gaussian primitives. We further design a dedicated gradient optimization scheme that ensures high reconstruction accuracy while drastically reducing resource demands. Evaluated on standard benchmarks, our method achieves up to 48% faster training, 12% lower memory consumption, and a 38.8% improvement in local resolution. This work establishes an efficient, scalable, and high-fidelity framework for cryo-EM structure determination.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Matrix & Tensor Methods
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

Cryo-electron microscopy (cryo-EM) has become a central tool for high-resolution structural biology, yet the massive scale of datasets (often exceeding 100k particle images) renders 3D reconstruction both computationally expensive and memory intensive. Traditional Fourier-space methods are efficient but lose fidelity due to repeated transforms, while recent real-space approaches based on neural radiance fields (NeRFs) improve accuracy but incur cubic memory and computation overhead. Therefore, we introduce GEM, a novel cryo-EM reconstruction framework built on 3D Gaussian Splatting (3DGS) that operates directly in real-space while maintaining high efficiency. Instead of modeling the entire density volume, GEM represents proteins with compact 3D Gaussians, each parameterized by only 11 values. To further improve the training efficiency, we designed a novel gradient computation to 3D Gaussians that contribute to each voxel. This design substantially reduced both memory footprint and training cost. On standard cryo-EM benchmarks, GEM achieves up to 48% faster training and 12% lower memory usage compared to state-of-the-art methods, while improving local resolution by as much as 38.8%. These results establish GEM as a practical and scalable paradigm for cryo-EM reconstruction, unifying speed, efficiency, and high-resolution accuracy. Our code is available at https://github.com/UNITES-Lab/GEM.

Problem

• Addressing computational and memory inefficiency in cryo-EM 3D reconstruction
• Overcoming fidelity loss from Fourier transforms in traditional methods
• Reducing cubic memory overhead of real-space neural reconstruction approaches

Innovation

• Uses 3D Gaussian Splatting for cryo-EM reconstruction
• Models proteins with compact 3D Gaussian representations
• Introduces efficient gradient computation reducing memory usage