Predicting amorphous drug solubility using molecular descriptors: an in-silico approach

Amorphous solubility defines the maximum achievable supersaturation and plays a critical role in drug bioavailability. In this study, in-silico models were developed to predict amorphous solubility for a dataset of 33 structurally diverse, poorly soluble drugs. Molecular descriptor-based models were evaluated using univariate and multivariate analyses, and crystalline solubility was incorporated to account for the thermodynamic contribution to amorphous solubility. Univariate analysis identified XLogP as a key predictor, while residual analysis indicated that certain compounds were less accurately predicted by XLogP alone, suggesting the need for multivariate models. Incorporation of crystalline solubility further improved predictive model performance ( \:R² = 0.76, Q 2 LOO = 0.70, and MSE LOO = 0.11). A consensus-based feature selection strategy within a multivariate regression framework identified three robust descriptors: XLogP, SpMax3_Bhv, and nHBint6, which were used to build predictive models achieving \:R² = 0.67, Q² LOO =0.55, and MSE LOO = 0.17. These findings demonstrate that amorphous solubility is governed by an interplay of lipophilicity, molecular topology, and hydrogen-bonding capacity. Multivariate models can provide mechanistic insight while enabling predictive assessment of amorphous solubility. This study highlights the potential of in-silico approaches to guide drug design and formulation development.