Design of cross-reactive antigens with machine learning and high-throughput experimental evaluation

Area of Science:

• Vaccinology
• Computational Biology
• Protein Engineering

Background:

• Antigen sequence variability in pathogens like HIV and influenza complicates vaccine design.
• Developing vaccines with broad protection requires antigens that elicit responses against multiple variants.
• Factor H binding protein (fHbp) from *Neisseria meningitidis* presents design challenges due to extensive mutant possibilities.

Purpose of the Study:

• To apply machine learning for designing improved vaccine antigens, specifically targeting fHbp.
• To overcome limitations in antigenicity data for training machine learning models.
• To engineer fHbp mutants that transfer specific epitopes while retaining broad reactivity.

Main Methods:

• Utilized computational models to predict fHbp properties.
• Employed Gaussian process (GP) machine learning to select promising and informative fHbp mutants.
• Experimentally evaluated selected mutants to refine the machine learning model iteratively.
• Performed biophysical and x-ray crystallographic characterization of the top mutant.

Main Results:

• Successfully designed fHbp mutants capable of transferring conformational epitopes from one fHbp variant to another.
• Maintained binding to cross-reactive epitopes while introducing new specificities.
• Confirmed structural integrity of the engineered fHbp mutant via biophysical and crystallographic analysis.

Conclusions:

• An integrated computational and experimental strategy can accelerate antigen design.
• This iterative platform holds potential for developing next-generation, broadly protective vaccines.
• Epitope engineering using machine learning is a viable approach for enhancing vaccine efficacy.