Al Hagbani Turki, Alzhrani Rami M, Algarni Majed Ahmed
Saudi Food and Drug Authority, Riyadh, Saudi Arabia.
Department of Pharmaceutics and Industrial Pharmacy, College of Pharmacy, Taif University, P.O. Box 11099, Taif, 21944, Saudi Arabia.
Sci Rep. 2025 Jul 12;15(1):25203. doi: 10.1038/s41598-025-10221-9.
Over the last decades, industrial employment of supercritical fluids (SCFs) as a trustworthy alternative of organic solvents has increased substantially. SCFs (mainly COSCF) have illustrated great potential to improve the solubility of poorly water-soluble drugs. Compared to traditional procedures, SCF-based processes possess irrefutable advantages like good sustainability, eco-friendliness, affordability, safety during application, thermodynamic stability, low consumption of energy and obtaining new products with better purity. This research was done with the aim of modeling the solubility of febuxostat (FBX) drug with the help of machine learning methods. Temperature and pressure are the input values on which the modeling is done. Regression models including GPR, KNN, and a voting regression model using these two basic models are employed in this research. As an innovative aspect of this research, in addition to the Voting model, the HHO algorithm has been used to tune the hyper-parameters of the models. The final models obtained were then evaluated and compared using different criteria. With the R criterion, the GPR predictive model has a score of 0.819 and the KNN model has a score of 0.854, but the voting model has a score of 0.980, which shows that the combined voting method using the other two models has a better result than both of them. Also, the RMSE error rate of the Voting model is 2.78 × 10 and with MAPE metric the error of this model is 3.81 × 10.
在过去几十年中，超临界流体（SCF）作为有机溶剂的可靠替代品在工业上的应用大幅增加。超临界流体（主要是COSCF）已显示出提高难溶性药物溶解度的巨大潜力。与传统方法相比，基于超临界流体的工艺具有不可辩驳的优势，如良好的可持续性、生态友好性、经济性、应用安全性、热力学稳定性、低能耗以及获得纯度更高的新产品。本研究旨在借助机器学习方法对非布司他（FBX）药物的溶解度进行建模。温度和压力是进行建模的输入值。本研究采用了包括高斯过程回归（GPR）、K近邻算法（KNN）以及使用这两种基本模型的投票回归模型在内的回归模型。作为本研究的创新点，除了投票模型外，还使用了哈里斯鹰优化（HHO）算法来调整模型的超参数。然后使用不同标准对获得的最终模型进行评估和比较。根据R准则，GPR预测模型的得分是0.819，KNN模型的得分是0.854，但投票模型的得分是0.980，这表明使用其他两个模型的组合投票方法比它们两者都有更好的结果。此外，投票模型的均方根误差（RMSE）率为2.78×10，使用平均绝对百分比误差（MAPE）指标时该模型的误差为3.81×10。
