Machine learning with API/polymer interaction mechanism: prediction for complex phase behaviors of pharmaceuticals and formulations

The high throughput prediction of the thermodynamic phase behavior of active pharmaceutical ingredients (APIs) with pharmaceutically relevant excipients remains a major scientific challenge in the screening of pharmaceutical formulations. In this work, a developed machine learning model efficiently predicts the solubility of APIs in polymers by learning the phase equilibrium principle and using a few molecular descriptors. Under the few-shot learning framework, thermodynamic theory (Perturbed-Chain Statistical Associating Fluid Theory, PC-SAFT) was employed for data augmentation, and computational chemistry was applied for molecular descriptors screening. The results showed that the developed machine learning model can predict the API-polymer phase diagram accurately, broaden the solubility data of APIs in polymers, and reproduce the relationship between API solubility and the interaction mechanisms between API and polymer successfully, which provided efficient guidance for the development of pharmaceutical formulations.