Format: ORAL

Lara M. Ksters1 Kevin Karbstein1 Martin Hofmann2 Ladislav Hodac1 Patrick Mder2,3,4 Jana Wldchen1,3

1 Biogeochemical Integration, Max-Planck-Institute for Biogeochemistry, Jena, 07745, Germany 2 Data-intesive Systems and Visualization Group (dAI.SY), Technical University, Ilmenau, 98693, Germany 3 German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, 04103, Leipzig, Germany 4 Faculty of Biological Sciences, Friedrich Schiller University, Jena, 07745, Germany

Integrative taxon-omics aims to combine genomics with a multitude of different complementary datasets such as morphology, ploidy level, information on hybridization and/or apomixis, geography, or ecology. Here, the goal is to advance species delimitation and to achieve a more reliable species classification. Machine learning (ML) encompasses a range of methods that autonomously learn to map a given input to a desired output. ML is rapidly gaining attention in the biological and botanical community for its ability to process highly multi-dimensional, large-scale datasets. However, there have been few attempts so far to combine an integrative approach with ML methodologies. We explore the potential of ML to classify species in the highly reticulate evolving Ranunculus auricomus complex, in which taxonomically challenging processes such as apomixis, hybridization, and polyploidy are prevalent. In our previous work, we assessed different fusion techniques with the integration of morphological data and genetic markers using convolutional neural networks (CNNs) that now serve as a basis for the fusion of extensive genetic data, morphology, ecology, and ploidy as well as reproductive information. We show that an integrative approach in ML species classification is not only feasible but that it increases the performance of the neural network. We hope to encourage a more holistic perspective of future ML models.