IsoNet2 determines cellular structures at submolecular resolution without averaging.

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Cryo-electron tomography (cryo-ET) is crucial for visualizing cellular structures.
• Reconstructing high-resolution 3D densities from cryo-ET data remains challenging due to noise and missing information.
• Existing methods often require averaging or extensive manual processing.

Purpose of the Study:

• To develop an end-to-end deep-learning method for direct, high-quality 3D density reconstruction from cryo-ET data.
• To improve resolution and enable atomic-level interpretation of cellular environments.
• To provide a user-friendly tool for researchers.

Main Methods:

• Introduced IsoNet2, a self-supervised deep-learning network.
• The unified network performs denoising, contrast transfer function correction, and missing-wedge restoration simultaneously.
• A graphical user interface (GUI) allows for dataset-specific fine-tuning.

Main Results:

• Achieved ~20 Å resolution reconstructions without averaging.
• Successfully resolved domain organization in HIV capsid proteins.
• Determined tRNA occupancy in individual ribosomes and in situ architectures of mitochondrial respiration complexes.

Conclusions:

• IsoNet2 enables high-quality 3D density reconstruction directly from cryo-ET data.
• The method facilitates atomic-level interpretation of complex cellular environments.
• IsoNet2 offers a powerful and accessible tool for structural and cellular biology research.