Identification of Monteverdia ilicifolia by fourier-transform mid-infrared spectroscopy associated with chemometrics and machine learning

Abstract

Monteverdia ilicifolia (Mart. Ex Reissek) Biral, a member of the Celastraceae botanical family, is widely recognized for its broad-spectrum therapeutic effects in South America, particularly in Brazil, where it is commonly referred as “espinheira-santa”. This study aimed to develop a chemometric and machine learning-based method for to accurately identify and differentiate M. ilicifolia from morphologically similar species used as adulterants. Fourier transform mid-infrared spectrometry (MIR-FTIR) was used to analyze leaves (n = 6 species, 3000 spectra), powders (n = 6 species, 3000 spectra) and extracts samples (n = 6 species, 600 spectra). The spectral datasets were predicted by Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA). The PLS-DA model was challenged with samples of other common plant species (n = 3) and commercial available M. ilicifolia (n = 10) to evaluate its predictive capability. PCA successfully distinguished between the plant species. PLS-DA achieved superior performance with extract samples, exhibiting sensitivity, specificity and accuracy of 94, 100 and 99 %, respectively. Machine learning algorithms were developed to better represent the leaves and powder samples through Random Forest and 10-fold validation methodology. The model yielded high accuracy in all sample types, with low false positive rate and excellent performance across the metrics of accuracy, recall, precision, F1 Score, Kappa index and Matthews Correlation Coefficient (MCC). PCA and PLS-DA models presented limitations over the complexity of leaves and powders samples. Machine learning algorithms showed robustness and flexibility, proving to be effective in the detection and discrimination of M. ilicifolia.