Chemists from Novo Nordisk A/S have initiated several optimisation projects using machine learning tools and robotics as alternatives to traditional design of experiments (DoE) methodologies and hands-on experiments. The reasoning is a pursuit to find better optima using fewer resources. Specifically, we have successfully applied Bayesian Optimization (BO) to a number of process optimizations in various areas such as synthetic chemistry, analysis development, stem-cell research, microbial cultivation, formulation development and others. BO is an iterative process optimization technique in which no prior data is required to get started. In essence, BO helps ensure that the experimentalists are performing the experiment from which we can learn the most (or which has the highest likelihood of being the optimal setting). We will present a few use cases including a chemical setup but also highlight the applicability of the code in other domains by showing the optimization of a recipe for an American brownie (a popular cake). Perspectives are drawn to our development of self-thinking robots that are able to autonomously optimize enzymatic transformations given a set of constraints and objectives. We are building an easy-to-use interface to Bayesian Optimization that will empower the domain experts to apply advanced machine learning while having little or no prior knowledge in coding. The code and user-interface are open-source.

Experiences in algorithm based chemical process optimization and early attempts towards closed loop automation

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.