The quantitative structure-property relationships for the gas-ionic liquid partition coefficient of a large variety of organic compounds in three ionic liquids

Ionic liquids (ILs) have unique properties as solvents and electrolytes, which need to be studied using innovative machine learning (ML) approaches and which allow the identification of a chemical environment that can be adapted to different applications. The gas-ionic liquid partition coefficients of organic compounds is one such application-oriented parameter for selecting both ionic liquids and organic compounds as quickly, cost-effectively, and as accurately as possible. Therefore, multiple linear regression (MLR) and random forest (RF) quantitative structure–property relationships (QSPRs) were developed for predicting the gas-ionic liquid partition coefficient (log K) of structurally variable organic solutes in the ionic liquids N-butyl-N-methylpyrrolidinium tris(pentafluoroethyl)trifluorophosphate ([BMPyrr]+[FAP]−), N-butyl-N-methylpyrrolidinium tricyanomethanide ([BMPyrr]+[C(CN)3]−) and 1-(2-methoxyethyl)-1-methylpyrrolidinium tris(pentafluoroethyl)trifluorophosphate ([MeoeMPyrr]+[FAP]−). All derived models have excellent prediction capability evidenced by high 5-fold cross-validated coefficients of determination in the range 0.88 – 0.94, complemented with other high statistical parameters. Compared to the MLR approach, the non-linear RF models statistics improved in two of three data series. Analysis of the molecular descriptors selected into MLR models revealed major solvent–solute interactions, with primary contributions from Coulomb and dipolar or hydrogen bonding interactions and followed by the descriptors that expose dispersion force related interactions. Relations to all the aforementioned solvent–solute interactions were also found in RF models descriptor interpretation. Comparison of models demonstrated that a common anion in different ILs produces a significant correlation between the data series log K values, while that of ILs with a common cation are less but still significantly correlated. The lower correlation could be attributed to varying structural differences in the corresponding ions, or the anion might have a more substantial role in determining partition properties with the organic solutes in the series.