Machine Learning-Based Assessment of Genetic Diversity and Population Structure in Hollyhock (Alcea rosea): A Comprehensive Morphological, Biochemical, and Molecular Approach

Hollyhock (Alcea rosea) is an important ornamental and medicinal plant. This study aimed to evaluate the genetic diversity and population structure of 28 endemic genotypes collected from various regions of Türkiye using an integrated approach combining morphological, biochemical, and molecular markers. Significant morphological variation was observed, and image analysis techniques were employed to classify flower and leaf colors. Biochemical assays revealed substantial diversity in bioactive compound levels, with total phenolic content ranging from 0.18 to 4.65 mg GAE/kg. Antioxidant activities also varied: FRAP (5.8 22.7%), CUPRAC (26.06–78.26%), and DPPH (13.6–70%). Molecular diversity was assessed using iPBS retrotransposon-based markers, with similarity indices ranging from 0.35 to 0.74, aligning with morphological and biochemical patterns. Principal Component Analysis (PCA) indicated dispersed distribution among genotypes in both two- and three dimensional plots, consistent with phenotypic variation. STRUCTURE analysis identified four distinct subpopulations based on genetic structure. Furthermore, machine learning (ML) models-Multilayer Perceptron (MLP), Support Vector Machines (SVM), and Random Forest (RF)-were employed to predict key biochemical and morphological traits. Model performance varied, with different R² values obtained for antioxidant related traits. This is the first comprehensive study of Alcea rosea integrating molecular data and ML tools to characterize genetic diversity and trait prediction.