Adaptive Gaussian representation for differentiable cryo-electron tomography reconstruction.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cryo-electron tomography (cryo-ET) offers 3D visualization of native biological structures.
• High-fidelity tomogram reconstruction is hindered by low signal-to-noise ratios and limited angular sampling.
• Existing methods struggle with artifacts like the missing wedge.

Purpose of the Study:

• To introduce CryoETGS, a novel differentiable learning framework for cryo-ET.
• To enable efficient and interpretable tomogram reconstruction using adaptive 3D Gaussian representations.
• To improve the fidelity and mitigate artifacts in cryo-ET reconstructions.

Main Methods:

• Developed CryoETGS, a differentiable learning framework utilizing adaptive 3D Gaussian representations.
• Implemented a hardware-accelerated differentiable rendering pipeline aligned with cryo-ET imaging geometry.
• Incorporated hierarchical initialization, adaptive densification, and tilt-weighted optimization.

Main Results:

• CryoETGS achieves state-of-the-art reconstruction performance on simulated and experimental datasets.
• The framework effectively mitigates missing wedge artifacts.
• Demonstrated high computational efficiency and real-time projection synthesis capabilities.

Conclusions:

• CryoETGS provides an efficient and interpretable method for cryo-ET data reconstruction.
• The adaptive 3D Gaussian representation significantly enhances tomogram fidelity.
• CryoETGS represents a significant advancement in cryo-ET data processing and analysis.