Visualizing Conformational Space of Functional Biomolecular Complexes by Deep Manifold Learning.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cellular functions rely on dynamic biomolecular complexes with diverse conformational states.
• Understanding these states is crucial for functional mechanisms and drug discovery.
• Current methods struggle to resolve the full conformational landscape.

Purpose of the Study:

• To introduce AlphaCryo4D, a deep manifold learning framework for atomic-level cryo-EM reconstructions.
• To visualize the conformational space and dynamics of biomolecular complexes.
• To improve 3D classification accuracy and reconstruction resolution.

Main Methods:

• Integration of 3D deep residual learning with manifold embedding of pseudo-energy landscapes.
• Utilizing an energy-based particle-voting algorithm for improved classification and resolution.
• Application to simulated and experimental datasets of proteasome, ribosome, and spliceosome.

Main Results:

• AlphaCryo4D achieved three times higher 3D classification accuracy than alternative methods on simulated data.
• Reconstructed continuous conformational changes of a 130-kDa protein at sub-3 Å resolution.
• Demonstrated generality in exploring hidden conformational states of macromolecular complexes.

Conclusions:

• AlphaCryo4D enables atomic-level visualization of biomolecular conformational continua.
• The approach facilitates the study of transient states and nonequilibrium dynamics.
• Potential for therapeutic discovery against highly dynamic biomolecular targets.