Integrative taxon-omics and machine learning to decipher reticulate plant speciation – a proof-of-concept from Thymus (Lamiaceae)

• Applicants: Dr. Kevin Karbstein; Dr. Jana Wäldchen
• Subject Area: Evolution and Systematics of Plants and Fungi
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 564834741

Modern integrative taxonomy combines 21st century high-throughput genomics and other complementary data sources (e.g., morphology, ploidy, reproduction, or ecology), and has started to become the new gold standard for species delimitation. It has also raised the awareness that what we call species can be ill-founded entities because of morphology-based, regional species descriptions. This is particularly true for taxonomically complex groups characterized by young origins, hybridization, or polyploidy. Here, the challenges of modern taxonomy become apparent: lack of appropriate analytical tools for intricate evolutionary processes or highly subjective ranking and manual fusion of datasets. Now, integrative taxonomy combined with machine learning (ML) enables standardized feature learning and data fusion to reduce subjectivity in species delimitation but also the ability to learn and handle intricate evolutionary processes. Here, we will use the economically important plant genus Thymus, which comprises hundreds of di- to polyploid, partly hybridogenous taxa distributed across Eurasia, with unknown evolutionary relationships and species status. We will perform whole-genome resequencing (WGR) based on DNA extracts from herbarium specimens, focusing on widespread diploid progenitors and important polyploid derivatives. Genome-wide SNPs, nuclear genes and plastome sequences, as well as ploidy, morphological, and ecological data will be analyzed using classical model-based approaches to clarify species boundaries. Nevertheless, this is prone to species over-splitting (e.g., DNA) or lumping (e.g., morphology), and is usually integrated on a manual basis. We will therefore also analyze Thymus datasets using state-of-the-art, fusion-based ML delimitation approaches based on specific pretraining rounds and learning of reticulate evolutionary relationships. ML results will be compared with classical DNA and integrative species delimitation results to obtain different views on species boundaries in relation to reticulate evolutionary processes and to make final taxonomic treatments. The DFG project aims to improve species delimitation by making it more integrative, objective, and suitable for reticulate evolution, while also accelerating the study of plant biodiversity in times of global change.

• DFG Programme: Research Grants