Czech J. Genet. Plant Breed., X:X | DOI: 10.17221/18/2026-CJGPB

Quantum-Inspired Machine Learning for Screening PEG-Induced Drought Stress Responses in Caraway (Carum carvi L.)Original Paper

Samuel Goitom Misginna ORCID...¹, Omar Gaoua ORCID...¹, Petra Röszlerová ORCID...², Musab A Isak ORCID...³, Ondrej Hejna ORCID...¹, Patrick Kamulegeya ORCID...¹, Eva Jozová ORCID...¹, Vladislav Čurn^1*

¹ Department of Genetics and Biotechnology, Faculty of Agriculture and Technology, University of South Bohemia, České Budějovice, Czech Republic

² Institute of Laboratory Diagnostics and Public Health, Faculty of Health and Social Sciences, University of South Bohemia, Czech Republic

³ Department of Agricultural Sciences and Technology, Faculty of Graduate School of Natural and Applied Sciences, Erciyes University, Kayseri, Türkiye

Drought is a significant factor limiting the growth and early establishment of caraway (Carum carvi L.), a valuable medicinal and aromatic plant. In this study, polyethylene glycol (PEG-6000)-induced osmotic stress assays were combined with statistical and machine learning (ML) approaches to assess early drought responses in five caraway cultivars and breeding materials. Seeds were subjected to four PEG concentrations (0, 5, 10 and 15 %), and key germination and seedling traits, including germination percentage (GP), root length (RL), root fresh weight (RFW), root dry weight (RDW), shoot height (SH), shoot fresh weight (SFW), and shoot dry weight (SDW), were measured. Higher PEG levels caused a sharp, accession-dependent decline in all traits, with germination dropping by 68 % at a 15 % PEG. Cultivars Aprim and H1b2/12 consistently showed better germination, shoot height, and biomass retention across stress levels, while Aklei exhibited lower germination but relatively stronger root growth, suggesting a differential adaptive response under osmotic stress. A linear model (LM) incorporating PEG concentration, accession, and their interaction served as the primary interpretable framework, explaining a large proportion of trait variation (R² = 0.81–0.94). Principal component analysis (PCA) and correlation analyses further revealed coordinated responses among biomass-related traits and differentiation in early-stage stress responses among accessions. Traditional ML models (MLP and SVR) were compared with quantum-inspired architectures (QiMLP and QiSVR); the quantum-inspired models showed comparable predictive performance in this dataset for certain traits, with QiMLP achieving the highest overall accuracy (R² = 0.88–0.94). This study presents an integrated phenotyping framework combining controlled stress assays with interpretable statistical modelling to evaluate early growth responses to PEG-induced drought stress in caraway. Overall, the results highlight accession-specific differences in early drought response and provide a useful basis for phenotyping and early-stage screening in caraway breeding.

Keywords: medicinal, aromatic and spice plants; neural networks; osmotic stress; predictive phenotyping; seedling traits

Received: February 10, 2026; Revised: May 8, 2026; Accepted: May 11, 2026; Prepublished online: May 26, 2026