Intelligence modeling of pharmaceutical solubility in supercritical CO2: Polynomial regression, extra trees, and Huber regression models and validation

Abstract

This research study investigates the relationship between temperature, pressure, and the properties of Erlotinib hydrochloride drug using a dataset comprising temperature, pressure, SC-CO₂ density, and mole fraction. Indeed, the solubility of medicine in the solvent was investigated via several models. The objective is to predict the density of solvent and the mole fraction of medicine. Three models, namely Extra Trees (ERT), Huber Regression (HBR), and Polynomial Regression (PR) are employed, and Sequential Model-Based Optimization (SMBO) is applied for hyperparameter tuning. Results showed promising outcomes, with Polynomial Regression achieving high accuracy in modeling both SC-CO₂ density and mole fraction. For SC-CO₂ density, Polynomial Regression achieves a significant R² score of 0.9973, RMSE of 8.4072E + 00, MSE of 7.0682E + 01, and MAE of 5.1116E + 00. In terms of mole fraction prediction, Polynomial Regression achieves an R² score of 0.9919, RMSE of 6.3569E-02, MSE of 4.0410E-03, and MAE of 5.7273E-02. The utilization of SMBO further enhances the performance of the models, demonstrating the effectiveness of this optimization technique. These findings have significant implications for understanding and utilizing the properties of SC-CO₂ in diverse scientific and industrial applications.