Assessment of lichen classification using metabolic fingerprint of Fourier transform infrared spectra

Lichens exhibit an intriguing diversity of fungi with approximately 20,000 accepted species worldwide. Their identification relies on time-consuming investigation of morphological and chemical characters in addition to phylogenetic evidence. The most widely utilized method for lichen chemotaxonomic investigations, thin layer chromatography, uses hazardous chemicals, while attenuated total reflectance Fourier transform infrared spectra provide simple, rapid, and environmentally friendly metabolic analysis. With the 12 different spectral pre-processing pipelines tested on data subsets of different sizes, of largest encompassing 12 families, 25 genera, and 35 lichenized fungal species from the full dataset of 25 families, 98 genera and 324 species, we found asymmetric least squares for baseline correction, and vector and z-score normalization as the best tools for accurate lichen identification. We conducted Random Forest classification on five mid-infrared spectral regions using the optimum pipeline for each subset, and the results indicate that taxonomically most informative FTIR spectra region was 1800-650 cm^-1, which we report here as the "lichen fingerprint region". The RF prediction accuracies across the taxonomic units tested using the fingerprint region were higher for species clustering (on average 87%), compared to genus (77%) or family (77%). The spectral regions of highest importance for species clustering are usually associated with cell wall carbohydrate and amide, which may provide characteristic features for lichen classification. Due to the high variability of FTIR spectra in the fingerprint region, the methodology that we propose here could be applied for studies on purified lichen metabolites, specific genera and ecological research questions, and quantitative analysis.