A Suite of Designed Protein Cages Using Machine Learning Algorithms and Protein Fragment-Based Protocols

Area of Science:

• Biotechnology and biomaterials science
• Computational biology and structural biology

Background:

• Designing self-assembling protein cages for biotechnological and medical applications presents significant challenges.
• Existing methods for protein cage creation are often unpredictable and difficult to control.

Approach:

• Employed a protein fragment-based approach for generating docked poses.
• Utilized and compared computational protocols for de novo interface design, highlighting the effectiveness of the machine learning program ProteinMPNN.
• Investigated the correlation between fragment-based sequence preferences, ProteinMPNN sequence inference, and experimental success.

Key Points:

• Machine learning, specifically ProteinMPNN, demonstrated increased experimental success in protein cage design.
• Agreement between fragment-based sequence preferences and ProteinMPNN predictions correlated with successful experimental outcomes.
• Experimental testing of larger, more polar interfaces provided insights into designing polar interactions.
• Achieved atomic resolution (2.0 Å) for five designed protein cage structures using X-ray crystallography and cryo-electron microscopy (cryo-EM).
• Reported structures of two incompletely assembled cages, offering insights into assembly failure mechanisms.

Conclusions:

• The study presents a new suite of computationally designed, tetrahedrally symmetric protein cages.
• The developed methodologies and reported structures substantially expand the library of protein nanoparticles and advance their design strategies.
• These findings pave the way for more predictable and successful creation of protein cages for diverse applications.