Systematic multi-trait AAV capsid engineering for efficient gene delivery.

Area of Science:

• Biotechnology
• Molecular Biology
• Gene Therapy

Background:

• Developing gene therapies requires efficient and manufacturable viral vectors like adeno-associated virus (AAV).
• Current AAV capsid library screening methods are inefficient for identifying vectors with multiple essential clinical traits.

Purpose of the Study:

• To present Fit4Function, a generalizable machine learning (ML) approach for systematically engineering multi-trait AAV capsids.
• To develop a method for creating AAV vectors with enhanced manufacturability and specific cellular targeting for clinical applications.

Main Methods:

• Leveraged a capsid library for uniform sampling of the manufacturable sequence space to generate reproducible screening data.
• Trained six sequence-to-function ML models using in vivo (mouse) and in vitro (human) data.
• Combined ML models to design and validate a multi-trait (liver-targeted, manufacturable) AAV capsid library.

Main Results:

• 88% of the designed multi-trait AAV capsid library variants met all six predetermined criteria.
• ML models accurately predicted AAV capsid biodistribution in macaques using data from mice and humans.
• Top AAV candidates showed high production yields, efficient murine liver transduction, and significantly enhanced human hepatocyte transduction (up to 1000-fold).

Conclusions:

• The Fit4Function strategy enables prediction of cross-species traits for peptide-modified AAV capsids.
• This ML approach is crucial for developing an atlas to predict AAV capsid performance across numerous traits, advancing gene therapy vector development.