| Autor: |
Beyramysoltan S; Department of Chemistry, University at Albany, State University of New York, 1400 Washington Avenue, Albany, New York 12222, United States., Chambers MI; Department of Chemistry, University at Albany, State University of New York, 1400 Washington Avenue, Albany, New York 12222, United States., Osborne AM; Department of Chemistry, University at Albany, State University of New York, 1400 Washington Avenue, Albany, New York 12222, United States., Ventura MI; Department of Chemistry, University at Albany, State University of New York, 1400 Washington Avenue, Albany, New York 12222, United States., Musah RA; Department of Chemistry, University at Albany, State University of New York, 1400 Washington Avenue, Albany, New York 12222, United States. |
| Abstrakt: |
The widespread abuse of "legal high" psychoactive plants continues to be of global concern because of their negative impacts on public health and safety. In forensic science, a major challenge in controlling these substances is the paucity of methods to rapidly identify them. We report the development of the D atabase o f P sychoactive P lants (DoPP), a new user-friendly tool featuring an architecture for the identification of plant unknowns, and the necessary regression statistics for the development and validation of psychoactive compound quantification. The application relies on the knowledge that terrestrial plants exhibit species-specific chemical signatures that can be revealed by direct analysis in real time─high-resolution mass spectrometry (DART-HRMS). Subsequent automated machine learning processing of libraries of these spectra enables rapid discrimination and species identification. The chemical signature database includes 57 available plant species. The rapid acquisition of mass spectra and the ability to sample the materials in their native form enabled the generation of the vast amounts of spectral replicates required for database construction. For the identification of sample unknowns, a data analysis workflow was developed and implemented using the DoPP tool. It utilizes a hierarchical classification tree that integrates three machine learning methods, namely, random forest, k-nearest neighbors, and support vector machine, all of which were fused using posterior probabilities. The results show accuracies of 98 and 99% for 10-fold cross-validation and external validation, respectively, which make the classification model suitable for identity prediction of real samples. |
