Abstract
BACKGROUND AND PURPOSE: Prediction of metabolism and solubility of tablet-form drugs is essential in pharmaceutical development, impacting drug efficacy, safety and formulation strategies. This study aimed to develop predictive models for classifying drugs according to metabolism and solubility within the Biopharmaceutical Drug Disposition Classification System. EXPERIMENTAL APPROACH: A dataset of 220 tablet-form drugs characterized by eleven molecular descriptors was analysed. The Kruskal-Wallis test identified relevant descriptors for metabolism (extensive vs. poor) and solubility (high vs. low) classifications. Probabilistic Neural Networks were employed for predictive modelling, with model parameters optimized to enhance accuracy. KEY RESULTS: Six molecular descriptors (hydrogen bond acidity, logarithm of the partition coefficient, distribution coefficient, hydrogen bond acceptor count, molecular weight and polar surface area) predicted metabolism class with 97 % accuracy. For solubility classification, five descriptors (dipolarity/polarizability, logarithm of the partition coefficient, distribution coefficient, hydrogen bond donor count and molecular weight) achieved 88 % accuracy. Removal of key descriptors significantly reduced model performance, confirming their importance. CONCLUSION: The developed models demonstrate robust predictive capability for drug metabolism and solubility classes as defined by the Biopharmaceutical Drug Disposition Classification System, supporting early-stage drug screening based solely on molecular structure. The lower accuracy observed for solubility prediction reflects its complex and multifactorial nature, highlighting the need for further refinement of molecular descriptors. These findings advance the field by providing effective computational tools to predict key biopharmaceutical properties, potentially accelerating the drug development process.