Exploring the Solubility Limits of Edaravone in Neat Solvents and Binary Mixtures: Experimental and Machine Learning Study

Abstract

This study explores the edaravone solubility space encompassing both neat and binary dissolution media. Efforts were made to reveal the inherent concentration limits of common pure and mixed solvents. For this purpose, the published solubility data of the title drug were scrupulously inspected and cured, which made the dataset consistent and coherent. However, the lack of some important types of solvents in the collection called for an extension of the available pool of edaravone solubility data. Hence, new measurements were performed to collect edaravone solubility values in polar non-protic and diprotic media. Such an extended set of data was used in the machine learning process for tuning the parameters of regressor models and formulating the ensemble for predicting new data. In both phases, namely the model training and ensemble formulation, close attention was paid not only to minimizing the deviation of computed values from the experimental ones but also to ensuring high predictive power and accurate solubility computations for new systems. Furthermore, the environmental friendliness characteristics determined based on the common green solvent selection criteria, were included in the analysis. Our applied protocol led to the conclusion that the solubility space defined by ordinary solvents is limited, and it is unlikely to find solvents that are better suited for edaravone dissolution than those described in this manuscript. The theoretical framework presented in this study provides a precise guideline for conducting experiments, as well as saving time and resources in the pursuit of new findings.

Keywords: COSMO-RS; deep learning; edaravone; green solvents; hyperparameter tuning; learning curve analysis; solubility.

Figure 1

Results of data curation of EDA solubility in neat methanol and ethyl acetate using values measured by [a] Li et al. [21] and [b] Qiu et al. [26]. The consensus lines characterize fitting to the van’t Hoff equation and black diamonds define solubility data included in the final dataset.

Figure 2

Results of data curation of EDA solubility in exemplary binary solvents using values measured by Li et al. [21]. The consensus lines characterize fitting to the Jouyban–Acree equation. The gray and black colors of markers and lines are used to distinguish lower from higher temperatures. The x_E, x*_MeOH, and x*_EtAc symbols denote the mole fraction solubility of EDA, the mole fraction of methanol, and the mole fraction of ethyl acetate in solute-free solutions, respectively.

Figure 3

Graphical representation of mole fraction solubility of edaravone in selected polar aprotic solvents. Gray and open symbols represent measured values and crosses depict values cured using the three-parameter van’t Hoff model.

Figure 4

Mole fraction solubility of edaravone at 25 °C in aqueous binary mixtures of selected polar diprotic solvents. Gray symbols represent measured values and crosses depict values cured using the JA model.

Figure 5

Results of regression models’ selection based on the distributions of the area under the AUC curve (blue dots) determined from learning curve analysis, loss values of test, and validation sets. Set A comprises the following five models: NuSVR, SVR, CatBoostRegressor, XGBRegressor, and HistGradientBoostingRegressor. In set B, twelve additional regressors were categorized including GaussianProcessRegressor, BaggingRegressor, RandomForestRegressor, LGBMRegressor, MLPRegressor, LassoLars, LassoLarsCV, Ridge, KNeighborsRegressor, AdaBoostRegressor, OrthogonalMatchingPursuitCV, and TransformedTargetRegressor.

Figure 6

Graphical illustration of the NuSVR regression model’s performance. The panels (a), (b), and (c) document the correlation between computed and consensus solubility values with annotation of the standard deviation as circle’s radius, applicability domain plots, and the results of learning curve analysis concerning both R² and MAE, respectively. The x_Eest symbol denotes the estimated EDA solubility values.

Figure 7

The experimentally and theoretically determined solubility values of EDA.

Figure 8

The results of optimization of the ΔC_p value for solubility computations using the COSMO-RS approach.