Enabling Data-Driven Solubility Modeling at GSK: Enhancing Purge Predictions for Mutagenic Impurities

In the pharmaceutical industry, solubility is a critical parameter influencing various stages of drug development, from early discovery to commercial manufacturing. This work showcases a high-throughput solubility screening workflow and describes the steps required to standardize and curate data suitably to allow automated data flow. Using the high-quality data, we developed a quantitative structure–property relationship model using gradient boosting and molecular descriptors, requiring only a 2D molecular structure to generate predictions. The accuracy of the model is competitive with alternative approaches where additional physical data is not required. A key use case for solubility predictions made in this way is in developing control strategies for mutagenic impurities, allowing for a data-driven and consistent method for calculating the solubility contribution to purge calculations. Further perspective is given on the future of the application of the model as a solubility prediction algorithm and on the approach to data-driven methodologies supporting drug development in general, highlighting the potential for federated learning approaches which use technological approaches to overcome the barrier to cross-industry data sharing.