High-throughput image-based seed phenotyping and multivariate analysis to characterize common bean ( Phaseolus vulgaris L.) accessions

Abstract

Common bean (Phaseolus vulgaris L.) seed phenotyping is essential for characterizing genetic diversity and identifying superior traits to support breeding for climate resilience and nutritional quality. Traditional manual techniques are increasingly being replaced by high-throughput, image-based digital phenotyping to ensure precision and efficiency in large-scale morphometric analysis. In this study, 30 common bean accessions from the Rural Development Administration (RDA) Gene Bank, South Korea, were phenotyped in 2025 using high-resolution image-based analysis to quantify key traits, including area, solidity, circularity, major/minor axis lengths, aspect ratio, and Feret diameter. One-way ANOVA revealed highly significant differences among accessions for all measured traits (p < 0.001), confirming substantial genotypic variability. Seed area ranged from 93.41 mm2 (IT160310) to 39.00 mm2 (IT337943), while roundness varied from 0.708 to 0.462, indicating pronounced morphological diversity. Spearman’s rank correlation showed a strong positive relationship between seed area and Feret diameter (r = 0.94), whereas aspect ratio and roundness exhibited a perfect negative correlation (r = −1.0). Hierarchical clustering and PCA effectively grouped accessions, with the first two components explaining 96.3% of total variation (PC1 and PC2). Validation against manual methods showed strong correlations (r = 0.95 for area; r = 0.94 for length), confirming ImageJ’s reliability. These findings provide a robust phenotypic foundation for breeding programs, enabling trait-based selection and supporting the integration of high-throughput pipelines into germplasm screening and future genomic studies, such as marker-trait association and genomic selection.

Plant phenotyping relevance

画像ベースで種子形態形質を高スループットに抽出し、手動測定との相関で検証しており、表現型取得手法が研究の中心である。

Code and data availability

The paper's seed phenotype dataset (30 common bean accessions, ImageJ-derived trait measurements) is not publicly deposited; the data availability statement says it is available only on reasonable request. No public code, images, or supplementary assets with an authors' URL are provided.

No evidence-backed public reproduction asset is currently recorded.