CryoDRGN-ET: deep reconstructing generative networks for visualizing dynamic biomolecules inside cells.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cryo-electron tomography (cryo-ET) enables visualization of macromolecules in native cellular environments at molecular resolution.
• Image processing remains a challenge for resolving structural heterogeneity in cryo-ET data.
• Existing methods struggle to capture diverse biomolecular states and conformations within cells.

Purpose of the Study:

• To introduce cryoDRGN-ET, a novel deep learning approach for heterogeneous reconstruction of cryo-ET subtomograms.
• To enable visualization of diverse macromolecular states and continuous motions in situ.
• To overcome limitations in current cryo-ET image processing for structural heterogeneity.

Main Methods:

• Developed cryoDRGN-ET, a deep generative model for 3D density map reconstruction directly from subtomogram tilt-series.
• Applied cryoDRGN-ET to analyze Mycoplasma pneumoniae ribosomes to validate translational state recovery.
• Utilized cryo-ET on cryo-FIB milled Saccharomyces cerevisiae cells to study in situ structures.

Main Results:

• Successfully recovered known translational states of M. pneumoniae ribosomes in situ.
• Revealed the structural landscape of S. cerevisiae ribosomes during translation.
• Captured continuous motions of fatty acid synthase complexes within S. cerevisiae cells.

Conclusions:

• CryoDRGN-ET effectively reconstructs heterogeneous macromolecular structures and dynamics from cryo-ET data.
• The method provides insights into the functional states and conformational heterogeneity of molecules in their native cellular context.
• This open-source software advances the field of in situ structural biology by addressing key image processing bottlenecks.