Model study and machine learning prediction of phase behavior of drug/polymer complex systems under different relative humidity

<p indent="0mm">To rationally design, prepare, and store pharmacological formulations, it is crucial to understand the phase behavior of complicated drug/polymer complexes under varying relative humidity conditions. By creating thermodynamic models of vapor-liquid and liquid-solid multiphase equilibria and Gordon-Taylor equation systems, the thermodynamic phase behavior and glass transition temperature of drug/polymer at different relative humidity (RH = 0%, 60%, and 75%) were studied in this paper. Combining five distinct machine learning algorithms with the perturbed-chain statistical associating fluid theory model parameters and molecular structure descriptors creates a novel approach for predicting the phase diagram of drug/polymer complex systems under different RH. The results demonstrate that the random forest algorithm has the highest prediction accuracy among machine learning techniques combined with thermodynamic model parameters and molecular structure descriptors. The determination coefficients <italic>R</italic>² of the test set results of drug solubility in polymer and glass transition temperature curves at different RH were 0.970 and 0.992, respectively. The study further shows that the machine learning algorithm coupled mechanism model parameters and molecular structure descriptors can achieve accurate prediction of drug/polymer phase diagrams under different RH, which is expected to provide effective guidance for rational design of drug preparations, preparation and storage conditions screening.