Adaptive Gaussian representation for differentiable cryo-electron tomography reconstruction
[Project Page]
Journal of Structural Biology (108281)
Chi Zhang
We present CryoETGS, a differentiable learning framework that reconstructs tomograms through adaptive 3D Gaussian representations of biological structures for cryo-ET.
Cryo-electron tomography (cryo-ET) enables 3D visualization of biological structures in their native state, but high-fidelity tomogram reconstruction remains challenging due to low signal-to-noise ratios and limited angular sampling. In this work, we present CryoETGS, a differentiable learning framework that reconstructs tomograms through adaptive 3D Gaussian representations of biological structures for cryo-ET. This representation enables efficient and interpretable reconstructions through a hardware-accelerated differentiable rendering pipeline aligned with the cryo-ET imaging geometry. CryoETGS incorporates hierarchical initialization, adaptive densification, and a tilt-weighted optimization strategy to enhance convergence and reconstruction fidelity. The framework further supports real-time projection synthesis and bidirectional conversion between voxel and Gaussian representations. Extensive experiments on both simulated and experimental datasets demonstrate that CryoETGS achieves state-of-the-art reconstruction results, effectively mitigates missing wedge artifacts, and exhibits high computational efficiency.
🔥 2025/12/17: The paper is accepted by the Journal of Structural Biology and is online now. More sources will be released as soon as possible.
2025/01/06: We released the basic project code.
# Clone the repository
git clone https://github.com/JachyLikeCoding/ETGS.git
cd CryoETGS
# Create environment
conda create -n etgs python=3.10
conda activate etgs
# Install dependencies
pip install -r requirements.txt
# Compile the CUDA rasterizer
pip install ./cryoETGS/submodules/simple-knn
pip install ./cryoETGS/submodules/diff-gaussian-rasterization
Organize your cryo-ET dataset in /data as follows:
<EMPIAR-XXXXX>
├── images/ # Tilt series (.mrc)
├── sparse/ # Ground truth tomogram (.mrc, reconstructed by WBP/SIRT/other methods)
└── tilt.rawtlt (tilt.aln) # Tilt angle (.tlt/.rawtlt/.aln)
To start the reconstruction process with default hyperparameters:
python train.py --config_name xxx
If you find CryoETGS useful in your research or refer to the provided baseline results, please star ⭐ this repository and consider citing 📝:
@article{zhang2025adaptive,
title={Adaptive Gaussian representation for differentiable cryo-electron tomography reconstruction},
author={Zhang, Chi and Yang, Zhidong and Han, Renmin and Zhang, Fa and Feng, Jieqing},
journal={Journal of Structural Biology},
pages={108281},
year={2025},
publisher={Elsevier}
}
For questions or issues, please open a GitHub issue or contact chizhang_cs@zju.edu.cn