Characterization And Optimization Of Biologics Manufacturing Using Space-filling Designs And Machine Learning

By Dogan Ornek, Nida Sayed, Matthew Sousa, Connor Crue, Ravi Chari, Stephanie Delzell, Adalia Cardoso, Roy Lin, and Philip J. Ramsey

Classical response surface designs concentrate data at the boundary of the experimental region, leaving the kinetic behavior inside it largely unobserved. This study tests an alternative on a recombinant AAV9 transient transfection process: a 24-run space-filling design analyzed with self-validating ensemble modeling, run in an Ambr 15 microbioreactor system. Six risk-assessed parameters were evaluated against four responses, including viral genome titer, capsid titer, percent full capsids, and transfection efficiency. The authors, from Boston Institute of Biotechnology and the University of New Hampshire, rank which parameters carry the most weight, establish operating ranges, and compare the results with earlier DoE and one-factor-at-a-time studies. Their findings on how those rankings differ by response may change how you think about trade-offs between titer and capsid quality.

Explore the full study to see the models, tolerance intervals, and methodology.