(655g) Fast-Tracking Process Optimization with Small Data and Machine Learning

Pharmaceutical process design and optimization is a time-consuming and complex task. Data-driven machine learning (ML) can accelerate the process by predicting behavioral trends, variable effects, and optimal conditions. However, conventional ML algorithms often require extensive datasets, posing a resource-intensive challenge in experimental data collection.

To address this bottleneck, we introduce an enhanced Bayesian optimization (BO) approach, enabling a reaction-agnostic pathway for designing and implementing intelligent experimental campaigns. SuntheticsML is an accessible online ML platform tailored for researchers without coding or ML expertise. The approach demonstrates compelling returns on material and experimental efficiency, as well as performance gains against a competitive baseline.

SuntheticsML is a versatile technology that allows numeric, discrete, and mixed-integer optimization problems with up to 20 input parameters, further facilitating bounded-target, multi-objective, and constrained-input optimizations. We will show case studies of SuntheticsML with industrial partners showcasing accelerated formulation optimization, process characterizations, and process development efforts in chemocatalytic reactions, biocatalytic cascades, crystallizations, in-vitro mRNA transcription processes, and more.

The in-lab validation of SuntheticsML convincingly demonstrates impressive returns on material efficiency, with up to a 75% reduction in the use of expensive or complex reagents. Experimental efficiencies enabling 9-12% increases in previously-optimized yields with a 2-32X reduction in optimization experiments.

The insights gained from this work redefine the landscape of reaction engineering, process development, and optimization, while simultaneously lowering barriers to the adoption of new chemical technologies.