Agro-morphological characterization and machine learning-based prediction of genetic diversity in six-row barley genotypes from Türkiye

Abstract The restricted genetic diversity observed in modern barley represents a significant obstacle to enhancing productivity. This study addresses this issue by characterising 445 six-row barley genotypes from the Osman Tosun Gene Bank in Türkiye. A comprehensive analysis of 22 agro-morphological traits, comprising 11 qualitative and 11 quantitative traits, was conducted to explore morphological, growth and phenological diversity. Principal component analysis identified four principal components, which collectively explained 72.86% of the total variance. Of these, the first two components accounted for 52.45%. Based on agro-morphological similarities, the genotypes were grouped into seven clusters. Clusters 5, 6, and 7 contained genotypes with high-yield traits, including early maturity, increased grain per spike, and higher thousand grain weight. The findings contribute directly to the expansion of the barley gene pool. Moreover, this study provides a comprehensive characterisation of the hitherto overlooked six-row barley germplasm present in Türkiye. This offers invaluable genetic resources for future breeding and molecular studies. Furthermore, the study compares the performance of three machine learning models (XGBoost, MARS, and Gaussian Processes) in predicting the harvest index from various traits. The XGBoost model demonstrated superior predictive ability, with the lowest RMSE (0.137), MAPE (0.222), and MAD (0.101) values, and was able to explain 99.8% of the barley variation. This research highlights the potential of machine learning algorithms in enhancing barley breeding by accurately predicting beneficial traits for yield improvement.