B. Dyulgerova
Agricultural Academy, Institute of Agriculture – Karnobat, 8400 Karnobat, Bulgaria

(Manuscript received 29 April 2026; accepted for publication 26 August 2026)

e-mail: bdyulgerova@abv.bg

Abstract. This study aimed to establish relationships among grain yield, protein content, and derived traits and indices, as well as the possibility of using near-infrared (NIR) spectroscopy for simultaneous selection of grain yield and protein content in barley breeding. A total of 50 genotypes of winter six-row barley were included in the study, consisting of 23 mutant lines derived from the Monika variety, 24 from the Izgrev variety, their parental varieties, and the standard control variety Veslets. Field experiments were conducted over three growing seasons (2021/2022-2023/2024) in the experimental field of the Institute of Agriculture-Karnobat, Southeastern Bulgaria. NIR spectra were acquired with a CropScan 1000B Near-Infrared Analyzer over the 720-1100 nm range at 10 nm intervals, and predictive models were built in R (v4.4.2) using a gradient boosting algorithm (XGBoost) with a pairwise ranking objective, evaluated through 5-fold cross-validation. Analysis of variance revealed significant effects of year, genotype, and their interaction on grain yield, protein content, and protein yield. The largest share of the variation in these traits was due to the conditions of the growing year. A strong negative correlation between yield and protein content was observed in the studied mutant lines. Grain protein deviation (GPD) was the most informative derived trait, reducing the dependence on yield and allowing more efficient identification of high-protein genotypes. In contrast, PY, GP&YD and EW showed strong redundancy with primary traits, resulting in minimal added information value. Predictive ability, assessed as the Spearman rank correlation between predicted and observed values, was strongly trait-dependent: protein content was predicted with high accuracy (R≈0.90-0.95), grain yield with moderate accuracy (R≈0.19-0.56), and GPD with intermediate but variable accuracy across years (R≈0.40-0.80), while the remaining composite indices showed weak and unstable predictive ability. The predictive performance based on NIR was highly dependent on the trait. Protein content showed high predictability, while grain yield demonstrated moderate predictive performance. GPD showed intermediate but variable predictability across years, while other composite indices were poorly predicted, likely due to poor spectral representation and indirect biological significance. NIR spectroscopy can be very useful for the rapid selection of grain quality traits, especially protein content, while its utility for yield and complex indices is limited and requires additional complementary data.