The Application of the Novel Expert Algorithm for Substance Identification (EASI) to Mass Spectra of Synthetic Cathinones

Description:
Novel psychoactive substances (NPS) such as synthetic cathinones continue to appear in forensic laboratory casework. Current mass spectral algorithms use an exemplar or consensus approach that often struggles to distinguish between compounds that have high structural and mass spectral similarities. The novel regression-based Expert Algorithm for Substance Identification (EASI), developed by our group, has successfully differentiated other NPS’s, such as fentanyl analogs, with false negative error rates less than ~9% compared to the consensus-based approach of ~26%. This project aims to extend EASI to structurally similar cathinones. Ten different 3,4-methylenedioxy cathinone isomers were analyzed in replicate using gas chromatography-electron-ionization mass spectrometry (GC-EI-MS) on at least 5 different instruments in two different laboratories. The replicate spectra were filtered and divided into a training set and a test set. Multivariate models were created from the training sets, and the validity of the models was assessed before applying them to the test sets. The EASI and consensus approaches were compared using conventional metrics of similarity and dissimilarity, such as mean absolute residuals, Euclidian distances, and dot products. Regardless of the chosen metric of spectral comparisons, binary classification using EASI outperformed the consensus with each metric with area under the receiver-operating characteristic curves (AUROC) of 1.000 for almost all 10 isomers. For EASI, the associated false negative rates were typically less than 0.5% at thresholds that provided zero false positives. With EASI, forensic chemists could have reduced rates of false-positive identifications while having increased confidence in their identifications and testimonies in court.

Speaker: Glen Jackson - West Virginia University