Comparison of species classification models of mass spectrometry data: Kernel Discriminant Analysis vs Random Forest; A case study of Afrormosia (Pericopsis elata (Harms) Meeuwen).

RATIONALE: The genus Pericopsis includes four tree species of which only Pericopsis elata (Harms) Meeuwen is of commercial interest. Enforcement officers might have difficulties discerning this CITES-listed species from some other tropical African timber species. Therefore, we tested several methods to separate and identify these species rapidly in order to enable customs officials to uncover illegal trade. In this study, two classification methods using Direct Analysis in Real Time (DART™) ionization coupled with Time-of-Flight Mass Spectrometry (DART-TOFMS) data to discern between several species are presented.

METHODS: Metabolome profiles were collected using DART™ ionization coupled with TOFMS analysis of heartwood specimens of all four Pericopsis species and Haplormosia monophylla (Harms) Harms, Dalbergia melanoxylon Guill. & Perr. Harms, and Milicia excelsa (Welw.) C.C. Berg. In total, 95 specimens were analysed and the spectra evaluated. Kernel Discriminant Analysis (KDA) and Random Forest classification were used to discern the species.

RESULTS: DART-TOFMS spectra obtained from wood slivers and post-processing analysis using KDA and Random Forest classification separated Pericopsis elata from the other Pericopsis taxa and its lookalike timbers Haplormosia monophylla, Milicia excelsa, and Dalbergia melanoxylon. Only 50 ions were needed to achieve the highest accuracy.

CONCLUSIONS: DART-TOFMS spectra of the taxa were reproducible and the results of the chemometric analysis provided comparable accuracy. Haplormosia monophylla was visually distinguished based on the heatmap and was excluded from further analysis. Both classification methods, KDA and Random Forest, were capable of distinguishing Pericopsis elata from the other Pericopsis taxa, Milicia excelsa, and Dalbergia melanoxylon, timbers that are commonly traded.