Phenotype-based core collections can improve the efficiency of germplasm management and evaluation by providing a smaller subset that preserves the principal structure of phenotypic variation. In this study, we developed a core collection from 332 rose accessions currently available in Korea through commercial and private channels. Multiyear phenotypic records were summarized at the accession level, and Shannon’s entropy (H) was calculated for each descriptor. Five alternative descriptor sets (D1–D5) were constructed by combining entropy-filtered qualitative traits with all quantitative traits. For each descriptor set, Gower distances were calculated and analyzed using principal coordinates analysis and unweighted pair-group method with arithmetic mean (UPGMA) clustering. A 15% core subset comprising 50 accessions was then selected through UPGMA cluster-based medoid selection. The resulting core subsets were evaluated based on mean deviation percentage (MDP) for quantitative traits, coverage ratio (CR%) across all descriptors, agreement with the clustering structure of the full set as measured by the adjusted Rand index (ARI) and normalized mutual information (NMI), and cophenetic correlation coefficients (CCC) for the UPGMA dendrograms. Among the five descriptor sets, D3—which comprised qualitative traits with H ≥ 0.7 together with all quantitative traits—provided the most favorable balance among the evaluation metrics. D3 showed low MDP, relatively high ARI and NMI, moderate CR%, and stable CCC values while reducing the number of phenotypic descriptors from 88 to 69. Thus, the resulting phenotype-based core subset of the Korean rose panel reduced both accession and descriptor numbers, thereby improving germplasm management and phenotypic evaluation.

Introduction
Materials and Methods
    Plant materials and experimental panel
    Phenotypic traits and data processing
    Shannon’s entropy and definition of descriptor sets
    Gower distance, principal coordinates analysisand clustering
    Core subset selection
    Evaluation of core subsets
    Statistical analyses and software
Results
    Phenotypic descriptors and Shannon’s entropy
    Structure of the Korean rose panel in PCoA andclustering space
    Core subsets obtained from alternative descriptorsets
    Properties of the D3-based core subset for theKorean rose panel
Discussion
Acknowledgement
References
Supplementary Data S1.
Supplementary Data S2.

• Seonghwa Bak(Department of Horticulture, Chonnam National University, Gwangju 61186, Korea)
• Myungshin Gwack(Department of Horticulture, Chonnam National University, Gwangju 61186, Korea)
• Tae-Ho Han(Department of Horticulture, Chonnam National University, Gwangju 61186, Korea, Interdisciplinary Program in IT-Bio convergence systems, Chonnam National University, Gwangju 61186, Korea) Corresponding author