Chemometric-assisted near-infrared spectroscopy (NIRS) for rapid biochemical profiling and diversity assessment in Job's tears (Coix lacryma-jobi)

Coix lacryma-jobi (Job’s tears) is a nutritionally rich yet underutilized cereal cultivated across Asia. However, a gap remains in large-scale profiling and the development of rapid analytical tools for screening. This study addresses that gap using an integrated multivariate and near-infrared spectroscopic (NIRS) approach. Evaluation of 119 accessions for 14 biochemical traits generated a robust reference dataset for both multivariate exploration of nutritional diversity and calibration and validation of NIRS models. Multivariate analysis (clustering, PCA, correlation) revealed distinct nutritional patterns among accessions, enabling differentiation and the identification of underlying nutritional structures. The observed patterns indicate the systematic variability in germplasm governed by combined biochemical responses. To develop NIRS models, spectral pre-processing techniques (e.g., SNV, detrending, Savitzky-Golay) enhanced signal quality and reduced baseline noise, enabling Partial Least Squares (PLS) regression. High predictive performance was achieved for protein (RSQ val = 0.95, RPD = 4.64), dietary fibre (RSQ val = 0.81, RPD = 2.32). Consistency was validated using a reliability test using Cronbach’s alpha (α = 0.52–0.98) and correlation coefficients ( r =0.48–0.97) . This high-throughput, non-destructive framework facilitates industrial-scale nutritional screening, empowering plant breeding for quality enhancement, compositional evaluation of raw materials for functional foods, and fermentation-linked applications. • Nutritional diversity of Coix germplasm accessed across 14 traits • Four nutritionally distinct groups and unique accessions identified • NIR spectra-based hierarchical clustering aligns with nutritional groupings • Near infrared spectroscopy prediction models achieved with R 2 >0.8 for multiple traits • High validation performance and reliability metrics confirm model robustness