Quantitative structure–property relationship (QSPR) modeling was investigated to predict drug and drug‐like compounds solubility in supercritical carbon dioxide. A dataset of 148 drug\drug‐like compounds, accounting for 3971 experimental data points (EDPs), was collected and used for modelling the relationship between selected molecular descriptors and solubility fraction data achieved by a nonlinear approach (Artificial neural network, ANN) based on molecular descriptors. Experimental solubility data for a given drug were published as a function of temperature and pressure. In the present study, 11 significant PaDEL descriptors (AATS3v, MATS2e, GATS4c, GATS3v, GATS4e, GATS3 s, nBondsM, AVP‐0, SHBd, MLogP, and MLFER_S), the temperature and the pressure were statistically proved to be sufficient inputs. The architecture of the optimized model was found to be {13,10,1}. Several statistical metrics, including average absolute relative deviation (AARD=3.7748 %), root mean square error (RMSE=0.5162), coefficient of correlation (r=0.9761), coefficient of determination (R2=0.9528), and robustise (Q2=0.9528) were used to validate the obtained model. The model was also subjected to an external test by using 143 EDPs. Sensitivity analysis and domain of application were examined. The overall results confirmed that the optimized ANN‐QSPR model is suitable for the correlation and prediction of this property.

The purpose of this work was to compare the performance of 7 meta‐heuristics algorithms namely: Dragonfly (DA), Ant Lion (ALO), Grey Wolf (GWO), Artificial Bee Colony (ABC), Particle Swarm (PSO), Whale (WAO), and a hybrid Particle Swarm with Grey Wolf (HPSOGWO) optimizers in terms of fine‐tuning hyper‐parameters of a hybrid quantitative structure property relationships (QSPR)‐support vector regression (SVR) for the prediction of molar fraction solubilities of drug compounds in supercritical carbon dioxide (SC‐CO2). A dataset of 168 drug compounds, 13 inputs, and 4490 experimental data points was used to achieve the goal. All 7 models were statistically and graphically approved while the HPSOGWO‐SVR was found to over‐perform with an average absolute relative deviation (AARD) of 0.706% and an AIC of −14,434,249. The model was subjected to an external test (validation) using 160 experimental data points that were not used in the training and the test set. The overall results proved that the obtained model has good predictivity ability and robustness.