Abstract Detail

Nº613/1185 - A machine learning approach to morphometry of Armeria (Plumbaginaceae) in peninsular Italy
Format: ORAL
Authors
Manuel Tiburtini1, Giovanni Astuti1, Fabrizio Bartolucci2, Liliana Bernardo3, Fabio Conti2, Duilio Iamonico4, Mauro Iberite4, Simone Orsenigo5, Luca Sandroni1, Lorenzo Peruzzi1
Affiliations
1 University of Pisa, Italy 2 University of Camerino, Italy 3 University of Calabria, Rende, Italy 4 Sapienza University of Rome, Italy 5 University of Pavia, Italy
Abstract
In the contemporary era of genome-wide phylogenomic analyses, the significance of morphology in Botany is crucial but often overlooked. Morphometry emerges as a method for scientifically and statistically handling morphological data, particularly in intricate groups like Armeria (Plumbaginaceae) in peninsular Italy, which has undergone multiple taxonomic revisions without yielding conclusive outcomes. In this study, we measured 27 features from 587 herbarium specimens across 34 populations spanning across the Apennines. The specimens encompass the A. arenaria, A. denticulata, A. canescens, and A. macropoda complexes for a total of 11 taxa, 10 of which putatively endemic to Italy. Since A. gracilis has been always traditionally related to the Balkan A. canescens, we also included the latter species for comparative purposes. To analyze the data, we employed two innovative machine learning approaches. Dimensionality reduction was achieved using a novel non-linear technique called UMAP (Uniform Manifold Approximation and Projection). Additionally, Gaussian mixture models (GMM) were used to test alternative grouping hypotheses. The 3D - UMAP analysis along the Italian peninsula reveals six main groups, i.e. from northern to southern Italy: A. arenariasubsp. praecox, A. arenaria subsp. marginata, A. denticulata complex (A. denticulata and A. saviana), two groups within the A. canescens complex (including also A. gracilis, A. garganica, and two populations referred as A. macropoda), andthe A. macropoda complex (A. aspromontana, A. brutia, A. macropoda s.str., and a population from the Pollino Massif referred as A. gracilis). Bayes factor calculations, fitting a supervised GMM, support the recognition of these six groups over other grouping hypotheses, constituting a comprehensive overview of our taxonomic exploration. Seed morphometric, karyological, and phylogenetic studies are ongoing and will be integrated into a conclusive taxonomic study.