LC–QTOFMS Presumptive Identification of Synthetic Cannabinoids without Reference Chromatographic Retention/Mass Spectral Information. II. Evaluation of a Computational Approach for Predicting and Identifying Unknown High-Resolution Product Ion Mass Spectra

Despite liquid chromatography–high-resolution tandem mass spectrometry (MS2) enables untargeted acquisition, data processing in toxicological screenings is almost invariably performed in targeted mode. We developed a computational approach based on open source chemometrics software that, starting from a suspected synthetic cannabinoid (SC) determined formula, searches for isomers in different new psychoactive substances web databases, predicts retention time (RT) and high-resolution MS2 spectrum, and compares them with the unknown providing a rank-ordered candidates list. R was applied on 105 SC measured data to develop and validate a multiple linear regression quantitative structure–activity relationship model predicting RT. Competitive Fragmentation Modeling for Metabolite Identification (CFM-ID) freeware was used to predict/compare spectra with Jaccard similarity index. Data-dependent acquisition was performed with an Agilent Infinity 1290 LC-6550 iFunnel Q-TOF MS with ZORBAX Eclipse-Plus C18 (100 × 2.1 mm2/1.8 µm) in water/acetonitrile/ammonium formate gradient. Ability of the combined RT/MS2 prediction to identify unknowns was evaluated on SC standards (with leave-one-out from the RT model) and on unexpected SC encountered in real cases. RT prediction reduced the number of isomers retrieved from a group of new psychoactive substances web databases to one-third (2,792 ± 3,358→845 ± 983) and differentiated between SC isomers when spectra were not selective (4F-MDMB-BUTINACA, 4F-MDMB-BUTINACA 2ʹ-indazole isomer) or unavailable (4CN-Cumyl-B7AICA, 4CN-Cumyl-BUTINACA). When comparing 30/40 eV measured spectra of 99 SC against RT-selected, CFM-ID predicted spectra of isomers, the right candidate ranked 1st on median and 4th on average; 54% and 88% of times the right match ranked 1st or within the first 5 matches, respectively. To our knowledge, this is the first case of extensive chemometrics application to toxicological screening. In most cases, presumptive identification (being based on computation, it requires further information for confirmation) of unexpected SC was achieved without reference measured information. This method is currently the closest possible to true unbiased/untargeted screening. The bottleneck of the method is the processing time required to predict mass spectra (ca. 30–35 s/compound using a 64-bit 2.50-GHz Intel® Core™ i5-7200U CPU). However, strategies can be implemented to reduce prediction processing time.