Automated multifactorial design of experiment and Bayesian optimisation algorithm approaches to method development...

During drug development, chromatography is frequently used for purity and stability testing of both drug substance and drug product. Reversed phase liquid chromatography (RPLC) is one of the most widely used methodologies due to its wide scope of application. In the later stages of drug development, the specified impurities and degradation products that define the critical quality attribute of the final API, also known as Key Predictive Sample Set (KPSS), are usually well defined and controlled. At this point, a method review enables selecting the most appropriate technique which should be the one providing optimal robustness (ICH-Q14[1]), with the support of Quality by Design (QbD) approaches. Supercritical Fluid Chromatography (SFC) is a preferred technique for its proven diversity in selectivity. The adoption of a technique which presents the most favourable environmental impact, such as, but not limited to, SFC, is also becoming increasingly important as laboratories strive to reduce carbon footprint. Re-developing a method requires high resource-demands in terms of staff, materials, and time. Any step of the process that can be automated can facilitate this approach, speeding up the delivery of the method whilst preserving robustness. In this article we describe how an SFC method was developed for the purity profiling of a late-stage oncology candidate, taking advantage of the superior selectivity of SFC towards structurally similar analytes, owed to the high orthogonality with R² as low as 0.014 towards the KPSS. We describe two approaches to automate the method development. Firstly, a multifactorial design of experiments (DoE) and secondly, an optimization via a Bayesian algorithm, which was completed in one night, highlighting the potential and limitations, with an insight into the robustness. Both methods achieved baseline separation with varying levels of automation embedded into the process and a large reduction of the resource demands when compared to traditional optimisation methods. Finally, we describe the beneficial environmental impact that implementing SFC methods can yield, with a calculated green score reduced to a value between 17 and 30 % compared to RPLC, depending on the number of runs per sequence.