ICECREAM: high-fidelity equivariant cryo-electron tomography.

Area of Science:

• Structural biology
• Microscopy
• Computational imaging

Background:

• Cryo-electron tomography (cryo-ET) is crucial for visualizing 3D cellular structures.
• Current deep learning methods struggle with low signal-to-noise ratios and the 'missing wedge' artifact in cryo-ET.
• Lack of ground truth data hinders algorithm development.

Purpose of the Study:

• To introduce ICECREAM, a novel deep learning framework for cryo-ET.
• To improve denoising and missing wedge correction in cryo-ET data.
• To provide a robust method applicable to various tomography problems.

Main Methods:

• ICECREAM integrates self-supervised learning with an equivariant imaging framework.
• The method utilizes two statistically independent views of the volume, obtainable via dose splitting or angular partitioning.
• It is designed to be broadly applicable to tomography datasets.

Main Results:

• ICECREAM demonstrates superior denoising performance compared to existing methods.
• It achieves more reliable correction of the 'missing wedge' artifact.
• The framework shows effectiveness across diverse experimental cryo-ET datasets.

Conclusions:

• ICECREAM significantly advances cryo-electron tomography data processing.
• The method offers improved accuracy and reliability in reconstructing 3D cellular architecture.
• ICECREAM provides a valuable tool for structural biology research.