Adaptive Machine Learning Framework for Optimizing the Affinity Purification of Adeno-Associated Viral Vectors.

Area of Science:

• Biotechnology
• Gene Therapy Manufacturing
• Machine Learning in Bioprocessing

Background:

• Adeno-associated viral (AAV) vectors are crucial for gene therapy, but their purification is a major bottleneck affecting cost and accessibility.
• Traditional purification methods are resource-intensive and often suboptimal for maximizing yield and quality.

Purpose of the Study:

• To develop and validate a machine learning framework for optimizing adeno-associated viral (AAV) vector purification using Bayesian optimization.
• To systematically refine affinity chromatography parameters to enhance AAV yield, purity, and transduction efficiency.

Main Methods:

• Implemented a closed-loop machine learning framework leveraging Bayesian optimization to refine affinity chromatography parameters (sample load, flow rate, media formulation).
• Applied the framework to purify clinically relevant AAV serotypes (AAV2, AAV5, AAV6, AAV9) using the AvXcel adsorbent.
• Iteratively optimized parameters over three or fewer cycles.

Main Results:

• Achieved significant improvements in AAV yield, increasing from a baseline of 70% to 97%-99% within three optimization cycles.
• Reduced host cell impurities by 230- to 400-fold across all tested serotypes.
• Consistently produced high-purity AAV vectors with preserved high transduction activity, demonstrating broad applicability.

Conclusions:

• The developed Bayesian optimization framework offers an efficient, data-driven approach to enhance AAV purification at the affinity capture step.
• This adaptive machine learning strategy accelerates process development, reduces manufacturing costs, and improves the accessibility of AAV-based gene therapies.
• Demonstrated transferability of purification data and process knowledge, highlighting the framework's robustness and potential for broad adoption in AAV manufacturing.