Design of Experiments (DOE) is a powerful and pragmatic tool for optimisation but it’s not the only way. Could there be even more powerful alternatives in the age of machine learning, AI and automation?

Active learning is an artificial intelligence method that makes use of machine learning algorithms to determine the next set of experiments to be carried out.

Traditional methods for process optimization based on DOE rely on domain specific expertise of the experimenter and are subjective in their interpretation and design. Active learning methodologies can be used to iterate following on from initial DOE based experimental design and promise more systematic iteration with less need for human input. For this reason they can be particularly powerful and can pave the way for automated closed loop optimisation.

We demonstrate here the application of an active learning based framework and an LLM with a cloud based digital experiment platform capable of planning and capturing results of experiments run on lab automation. We used it to optimize miniaturized fermentation for recombinant protein production in bacteria. We performed 4 iterations which resulted in a 5 fold increase in maximum rate of protein production and 2 fold increase in protein titre. We also attained many key insights into behavior.