Geometric morphometrics and machine learning challenge currently accepted species limits of the land snail Placostylus (Pulmonata: Bothriembryontidae) on the Isle of Pines, New Caledonia

Abstract

AbstractSize and shape variations of shells can be used to identify natural phenotypic clusters and thus delimit snail species. Here, we apply both supervised and unsupervised machine learning algorithms to a geometric morphometric dataset to investigate size and shape variations of the shells of the endemic land snail Placostylus from New Caledonia. We sampled eight populations of Placostylus from the Isle of Pines, where two species of this genus reportedly coexist. We used neural network analysis as a supervised learning algorithm and Gaussian mixture models as an unsupervised learning algorithm. Using a training dataset of individuals assigned to species using nuclear markers, we found that supervised learning algorithms could not unambiguously classify all individuals of our expanded dataset using shell size and shape. Unsupervised learning showed that the optimal division of our data consisted of three phenotypic clusters. Two of these clusters correspond to the established species Placostylus fibratus and P. porphyrostomus, while the third cluster was intermediate in both shape and size. Most of the individuals that were not clearly classified using supervised learning were classified to this intermediate phenotype by unsupervised learning, and most of these individuals came from previously unsampled populations. These results may indicate the presence of persistent putative-hybrid populations of Placostylus in the Isle of Pines.