Scholarlink

<p>The primary reason why we need to develop methods for accurately predicting the solubility of anticancer pharmaceuticals in supercritical carbon dioxide (scCO<sub>2</sub>) is the need for environmentally friendly pharmaceutical processes and advancements in drug delivery systems. Throughout this research, Random Forest and eXtreme Gradient Boosting (XGBoost) models were created to estimate the solubility of four different anticancer drugs (gefitinib, gemcitabine, letrozole, and tamoxifen) in scCO₂, across a wide range of thermodynamic conditions (temperatures = 308–348&#xa0;K; pressures = 120–400&#xa0;bar). We compiled a curated experimental dataset consisting of 100 measurements and split it into training (88 samples) and testing sets (12 samples). Out of these two models, XGBoost was far superior in terms of predictive ability compared to the Random Forest model. On the overall dataset, XGBoost achieved R² = 0.976, RMSE = 0.263&#xa0;g/L, and MAE = 0.216&#xa0;g/L. The Random Forest model achieved R² = 0.825, RMSE = 0.710&#xa0;g/L and MAE = 0.452&#xa0;g/L. Feature importance analysis of both procedures indicates that pressure-driven density effects are the main factors influencing solubility according to RF, with pressure at 43.9% and density at 40.7%. In contrast, XGBoost assigned more importance to temperature at 36.2%, reflecting additional nonlinear thermal interactions. By performing further uncertainty and sensitivity analyses, it is evident that the robustness of both models is established; pressure (93.4% sensitivity) and temperature (83.3% sensitivity) have the greatest effect on the stability of the predictive models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning for prediction of anticancer drug solubility in supercritical CO₂: a comparative study of random forest and XGboost

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Mostafa Khajeh,
  • Mansour Ghaffari-Moghaddam,
  • Bahareh Mahmoudi

摘要

The primary reason why we need to develop methods for accurately predicting the solubility of anticancer pharmaceuticals in supercritical carbon dioxide (scCO2) is the need for environmentally friendly pharmaceutical processes and advancements in drug delivery systems. Throughout this research, Random Forest and eXtreme Gradient Boosting (XGBoost) models were created to estimate the solubility of four different anticancer drugs (gefitinib, gemcitabine, letrozole, and tamoxifen) in scCO₂, across a wide range of thermodynamic conditions (temperatures = 308–348 K; pressures = 120–400 bar). We compiled a curated experimental dataset consisting of 100 measurements and split it into training (88 samples) and testing sets (12 samples). Out of these two models, XGBoost was far superior in terms of predictive ability compared to the Random Forest model. On the overall dataset, XGBoost achieved R² = 0.976, RMSE = 0.263 g/L, and MAE = 0.216 g/L. The Random Forest model achieved R² = 0.825, RMSE = 0.710 g/L and MAE = 0.452 g/L. Feature importance analysis of both procedures indicates that pressure-driven density effects are the main factors influencing solubility according to RF, with pressure at 43.9% and density at 40.7%. In contrast, XGBoost assigned more importance to temperature at 36.2%, reflecting additional nonlinear thermal interactions. By performing further uncertainty and sensitivity analyses, it is evident that the robustness of both models is established; pressure (93.4% sensitivity) and temperature (83.3% sensitivity) have the greatest effect on the stability of the predictive models.