Hallucination of closed repeat proteins containing central pockets

Area of Science:

• Protein design
• Computational biology
• Biophysics

Background:

• Pseudocyclic proteins feature repeated subunits forming a central cavity for ligand binding or enzymatic activity.
• Existing methods for designing such proteins are limited in scope and exploration of design space.

Purpose of the Study:

• To develop a deep-learning approach for exploring the design space of closed repeat proteins.
• To generate novel pseudocyclic protein designs with specified repeat numbers and lengths.

Main Methods:

• A deep-learning model was employed to design closed repeat proteins based on repeat number and length.
• 38 diverse pseudocyclic protein designs were produced in Escherichia coli.
• Biophysical characterization and X-ray crystallography were used to validate the designed structures.

Main Results:

• Biophysical data for 38 designs were consistent with computational models.
• Three crystal structures closely matched the designed protein architectures.
• The study demonstrated the feasibility of AI-driven protein design for pseudocyclic proteins.

Conclusions:

• The deep-learning approach enables broad exploration of closed repeat protein design space.
• The generated protein designs are structurally accurate and biophysically validated.
• These designs offer promising scaffolds for developing small-molecule binders and enzymes.