INTEGRATED NUMERICAL AND MIXED-DESCRIPTOR PHENOTYPING REVEALS COMPLEMENTARY STRUCTURE IN LENTIL (LENS CULINARIS MEDIK.) GERMPLASM

Abstract

Mixed-scale germplasm data can yield different biological groupings depending on whether continuous and coded descriptors are analyzed separately or jointly. We evaluated 518 lentil (Lens culinaris Medik.) accessions characterized by 10 continuous agro-morphological traits and 20 coded descriptors. Standardized continuous traits were analyzed by principal component analysis (PCA) followed by fuzzy c-means clustering, whereas continuous and coded descriptors were integrated through a factor-analysis-of-mixed-data-based embedding and a second fuzzy partition. Gower dissimilarity supported mixed-data validation, and the adjusted Rand index quantified agreement between pathways. Yield per plant was strongly related to pods per plant (r = 0.807) and biological yield per plant (r = 0.681). The first three principal components explained 67.35% of continuous-trait variance and resolved three profiles: a high-yield, high-pod group; a later, taller, lower-yield group; and an early, bold-seeded group. The mixed pathway resolved two broader descriptor-defined groups (fuzzy partition coefficient = 0.567; Xie-Beni index = 1.265; Gower silhouette = 0.087). Agreement between numerical and mixed allocations was moderate (adjusted Rand index = 0.325), showing that coded descriptor states altered accession grouping beyond the continuous-trait solution. High-membership accessions can represent group specific phenotypes, whereas transitional accessions may provide useful recombination material. The framework offers a transparent way to combine performance, morphology, and assignment uncertainty in lentil germplasm characterization.