Raloxifene solubility in supercritical CO2 and correlation of drug solubility via hybrid machine learning and gradient based optimization

Abstract

One of the problems with new medications is their poor water solubility that is possible to be addressed by using supercritical method. This study aims to predict the solubility of raloxifene and the density of supercritical CO₂ using temperature and pressure as inputs to analyze the supercritical processing for production of drug nanoparticles. Three regression models, Extra Trees (ET), Random Forest (RF), and Gradient Boosting (GB) were proposed and optimized using Gradient-based optimization to predict density and solubility of drug. In predicting the density of supercritical CO₂, GB attained an R² value of 0.986, reflecting an excellent agreement between its estimates and the true measurements. The model exhibited an RMSE of 23.20, indicating high accuracy, with a maximum error of 33.06. Regarding the solubility of raloxifene, the ET model yielded the highest R-squared score of 0.949, indicating a good fit to the data. The model exhibited an RMSE of 0.41, with a maximum error of 0.90. Comparatively, the RF and GB models obtained slightly lower precision, for the solubility of raloxifene. The RF model exhibited an RMSE of 0.55, while the GB model had an RMSE of 0.72. The optimized models were found to be reliable in predicting solubility and density within the supercritical processing field.

Keywords: Drug computation; Extra trees; Gradient boosting; Machine learning; Process modeling; Random forest; Solubility.

Fig. 1

Visualization of the drug dataset using Pair Plot evaluation.

Fig. 2

Gradient-based optimization of hyperparameters workflow.

Fig. 3

Actual and calculated Solubility of Raloxifene using three models.

Fig. 4

Actual and calculated CO₂ Density using three models.

Fig. 5

Individual Effect of P on Solubility of Raloxifene.

Fig. 6

Individual Effect of T on Solubility of Raloxifene.

Fig. 7

Individual Effect of P on CO₂ Density.

Fig. 8

Individual Effect of T on CO₂ Density.

Fig. 9

3D surface for Solubility of Raloxifene. The figure is drawn by Python v3.8 which can be freely downloaded from the link: https://www.python.org.

Fig. 10

3D surface for CO₂ Density. The figure is drawn by Python v3.8 which can be freely downloaded from the link: https://www.python.org.

Fig. 11

Feature importance for Density prediction.

Fig. 12

Feature importance for Solubility prediction.