Hallucinating symmetric protein assemblies.

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Deep learning generative models offer new avenues for exploring protein structure space.
• Current methods are limited to natural protein sequences and structures.

Purpose of the Study:

• To generate novel symmetric protein homo-oligomers using deep network hallucination.
• To explore the potential of deep learning in designing complex protein architectures.

Main Methods:

• Deep network hallucination was employed to generate protein structures.
• Specifications included the number of protomers and protomer length.
• Experimental validation using X-ray crystallography and cryo-electron microscopy.

Main Results:

• Seven designed protein homo-oligomers were crystallized, showing high structural similarity to computational models (median RMSD: 0.6 Å).
• Three giant ring structures (10-nm diameter) with up to 1550 residues and C33 symmetry were determined via cryo-EM.
• Generated structures significantly differ from previously known protein structures.

Conclusions:

• Deep learning enables the creation of diverse and novel protein structures.
• This approach opens possibilities for designing complex protein-based nanomachines and biomaterials.