Efficient discovery of new medicine formulations using a semi-self-driven robotic formulator

We present the discovery of new medicine formulations using a semi-self-driven robotic formulator. Solubilising drugs is a significant challenge in the pharmaceutical industry, with the majority of active molecules in development for therapies being poorly soluble. The discovery of high solubility drug formulations, required for efficacious medicines, is a highly complex challenge involving the mixing of active molecules with excipients in thousands of potential combinations. We have developed a self-driving laboratory process for the production, assessment, and optimisation of solubility of liquid formulations suitable for injectable medicines, and apply it to the example molecule curcumin. In a series of loops driven by Bayesian optimisation our system discovered 7 lead formulations with high solubility (> 10mg/mL) after sampling only 256 out of 7776 potential formulations (~3%) in only a few days. Our workflow involves the generation of a seed sample set determined by clustering, followed by subsequent formulation optimisation performed by a liquid handling robot. Beyond presenting an efficient workflow for the optimisation and discovery of new liquid formulations, this work forms the basis for a more generalised optimisation workflow that could be applied to any formulation problem in the future, including those where no prior information is known.