Machine learning approach to support taxonomic species discrimination based on helminth collections data

Parasites & Vectors

Flávio L. de Mello¹  Samanta C. C. Xavier²  Alena Mayo Iñiguez²  Victor Hugo Borba³  José Roberto Machado-Silva⁴  Coralie Martin⁵ 

[1] Departamento de Engenheira Eletrônica e Computação, Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil;Laboratório de Biologia de Tripanosomatídeos-LABTRIP, Instituto Oswaldo Cruz, IOC-FIOCRUZ, Rio de Janeiro, RJ, Brazil;Laboratório de Biologia de Tripanosomatídeos-LABTRIP, Instituto Oswaldo Cruz, IOC-FIOCRUZ, Rio de Janeiro, RJ, Brazil;Laboratório de Helmintologia Romero Lascasas Porto, Faculdade de Ciências Médicas, UERJ, Rio de Janeiro, RJ, Brazil;Laboratório de Helmintologia Romero Lascasas Porto, Faculdade de Ciências Médicas, UERJ, Rio de Janeiro, RJ, Brazil;Unité Molécules de Communication et Adaptation des Microorganismes (MCAM, UMR 7245), Muséum National d’Histoire Naturelle, CNRS, CP52, Paris, France;

关键词: Taxonomy;    Artificial intelligence;    Species identification;    Capillaridae;    Parasite eggs;   

DOI  :  10.1186/s13071-021-04721-6

来源: Springer

【 摘 要 】

BackgroundThere are more than 300 species of capillariids that parasitize various vertebrate groups worldwide. Species identification is hindered because of the few taxonomically informative structures available, making the task laborious and genus definition controversial. Thus, its taxonomy is one of the most complex among Nematoda. Eggs are the parasitic structures most viewed in coprological analysis in both modern and ancient samples; consequently, their presence is indicative of positive diagnosis for infection. The structure of the egg could play a role in genera or species discrimination. Institutional biological collections are taxonomic repositories of specimens described and strictly identified by systematics specialists.MethodsThe present work aims to characterize eggs of capillariid species deposited in institutional helminth collections and to process the morphological, morphometric and ecological data using machine learning (ML) as a new approach for taxonomic identification. Specimens of 28 species and 8 genera deposited at Coleção Helmintológica do Instituto Oswaldo Cruz (CHIOC, IOC/FIOCRUZ/Brazil) and Collection de Nématodes Zooparasites du Muséum National d’Histoire Naturelle de Paris (MNHN/France) were examined under light microscopy. In the morphological and morphometric analyses (MM), the total length and width of eggs as well as plugs and shell thickness were considered. In addition, eggshell ornamentations and ecological parameters of the geographical location (GL) and host (H) were included.ResultsThe performance of the logistic model tree (LMT) algorithm showed the highest values in all metrics compared with the other algorithms. Algorithm J48 produced the most reliable decision tree for species identification alongside REPTree. The Majority Voting algorithm showed high metric values, but the combined classifiers did not attenuate the errors revealed in each algorithm alone. The statistical evaluation of the dataset indicated a significant difference between trees, with GL + H + MM and MM only with the best scores.ConclusionsThe present research proposed a novel procedure for taxonomic species identification, integrating data from centenary biological collections and the logic of artificial intelligence techniques. This study will support future research on taxonomic identification and diagnosis of both modern and archaeological capillariids.Graphical abstract

【 授权许可】

CC BY