(515b) Machine Learning-Assisted Autonomous Reactor Characterization: Mass Transfer Coefficients for Oxidative Biocatalytic Reactions

Determination of mass transfer coefficients (kLa) is a critical requirement in drug development processes that require aerobic oxidations or other forms of reactions that are limited by supply of gaseous species to a liquid/slurry phase. kLa values depend on several parameters other than reaction conditions, such as reactor geometry, the number of baffles in a reactor, the position, size, and geometry of the agitator. As a result, design of an accurate empirical model for estimating kLa across scales remains very challenging. In addition, the overall kinetics of biocatalytic oxidation reactions are very sensitive to kLa values and thus a slight change in the kLa value across different scales can significantly increase the complexity of the process scale up due to lack of sufficient process information. Therefore, accurate determination of kLa across different scales is often a requirement for development of several drug synthesis processes.

The first part of this talk will focus on developing an accurate predictive model for determination of kLa values across scales ranging from 100 mL to 100 L scale. Specifically, the application of machine learning (ML) based algorithms on archived mass transfer characterization data for developing an accurate kLa prediction model, along with estimation of uncertainty in the ML model predictions will be discussed. The second part of the talk will focus on application of an active learning approach that enables a fully autonomous reactor characterization platform by leveraging ML-guided design of experiments for efficient exploration of the design space by executing minimal number of experiments.