Abstract
The application of Quantitative Structure–Property Relationship (QSPR) modeling to the prediction of reversed-phase liquid chromatography retention behavior of synthetic cannabinoids (SC), and its use in aiding the untargeted identification of unknown SC are described in this paper. 1D, 2D molecular descriptors and fingerprints of 105 SC were calculated with PaDEL-Descriptor, selected with Boruta algorithm in R environment, and used to build-up a multiple linear regression model able to predict retention times, relative to JWH-018 N-pentanoic acid-d5 as internal standard, under the following conditions: Agilent ZORBAX Eclipse Plus C18 (100 mm × 2.1 mm I.D., 1.8 μm) column with Phenomenex SecurityGuard Ultra cartridge (C18, 10 mm × 2.1 mm I.D., < 2μm) kept at 50°C; gradient elution with 5-mM ammonium formate buffer (pH 4 with formic acid) and acetonitrile with 0.01% formic acid, flow rate 0.5 mL/min. The model was validated by repeated k-fold cross-validation using two-third of the compounds as training set and one-third as test set (Q2 0.8593; root mean squared error, 0.087, ca. 0.56 min; mean absolute error, 0.060) and by predicting relative Retention Times (rRT) of 5 SC left completely out of the modeling study. Application of the model in routine work showed its capacity to discriminate isomers, to identify unexpected SC in combination with mass spectral information, and to reduce the length of the list of candidate isomers to ca. one-third, thus reducing significantly the time required for predicting high-resolution product ion spectra to be compared to the unknown using a computational Mass Spectrometry (MS) search/identification approach.
