Wissenschaftskolloquium

20.10.
Species delimitation 4.0: Using AI to uncover fine-scale plant phylogenomic and phenotypic variation
Dr. Kevin Karbstein, Friedrich-Schiller-Universität Jena

Deep learning (DL) is opening new possibilities for connecting the rapidly growing volume of digital biodiversity data with fine-scale questions in evolution and ecology. But can artificial intelligence capture the subtle genetic and morphological variation needed to understand and delimit closely related species? Using Ranunculus auricomus as a model, together with species groups such as Dactylorhiza (orchids) and Zanthoxylum (prickly ashes), we show how DL networks can automatically extract phylogenetic signals from single-copy nuclear gene sequences and identify species even under complex evolutionary scenarios. We combine dimensionality reduction and clustering with quantitative benchmarks against recently updated integrative taxonomies to evaluate the accuracy and reproducibility of DL-based species delimitation. We also explore morphology: in R. auricomus, DL networks automatically extract intricate leaf-shape variation from both field and herbarium images, revealing traits of clear taxonomic relevance that can be validated using geometric morphometrics. Looking ahead, these approaches point towards a shift from manually evaluating individual lines of evidence to automated data fusion for integrative taxonomy. By combining genomic and morphological resources from global databases such as NCBI, BOLD, or GBIF, DL provides a scalable and reproducible framework for discovering and revising plant diversity worldwide.
An outlook on new projects and applications will conclude the talk.

As usual, we will also broadcast via zoom for those who can not join in person: https://zoom.us/j/94510498226

Zum Online-Vortrag