Botanical authentication using one-class modeling

Authors: Harnly, James - Jim

Submitted to: Journal of AOAC International
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/11/2023
Publication Date: 2/8/2023
Citation: Harnly, J.M. 2023. Botanical authentication using one-class modeling. Journal of AOAC International. 106(4)1077–1086. https://doi.org/10.1093/jaoacint/qsad023.
DOI: https://doi.org/10.1093/jaoacint/qsad023

Interpretive Summary: One-class modeling is a simple way to authenticate a botanical material. Stated simplistically, a collection of authentic reference botanicals are analyzed using a non-targeted metabolomic method and used to construct a model which is used to determine if a test sample is similar or dissimilar (authentic or adulterated) to the reference samples. There is no need to specify an adulterant or to develop an adulterated test sample for comparison. The key aspect of the method is the use of a non-targeted metabolomic method (any chromatographic or spectral method) which provides a wide coverage of the signals of the components in a sample. Modeling is achieved using chemometric methods which can be designed to be very sensitive to differences in the chromatograms or spectra of the reference and test samples. One-class modeling is much less cumbersome than the previous Probability of Identification method validated by AOAC International.

Technical Abstract: Sample authentication using one-class modeling is simpler and more versatile than the approved AOAC Probability of Identification (POI) method. This approach develops a one-class model for the reference samples and does not identify or model non-authentic or adulterated samples. Unknown samples are classified as either authentic (in the same class as the reference samples) or not authentic (outside the reference model class). The authentication method uses flow injection mass spectrometry (FIMS) and chemometric analysis with pre-processing steps of sample vector normalization and autoscaling. Reference samples are selected from an inclusivity frame, collected, and analyzed by FIMS. Autoscaling reduces the impact of the relative ion intensities and provides much greater sensitivity to changes in the intensities of individual variables. Detection limits for each variable can be predicted based on an average spectrum and types of uncertainty (random or proportional). Method validation is readily achieved using cross validation. Detection of adulteration is spectrally oriented and requires no identification of adulterants. Thus, both the data acquisition and the selection of adulterants are untargeted.