Enhancing Predictive Ability of Agronomic and Quality Traits in Ethiopian Malting Barley ( Hordeum vulgare L.) Using Spectral Variable Selection Methods

Abstract

ABSTRACT Phenomic selection (PS) offers a cost‐effective , breeder‐friendly approach for public breeding programmes with limited access to genotyping or restricted financial resources for laboratory infrastructure. Since PS relies on high‐throughput phenotyping data, which is often derived from near‐infrared spectroscopy (NIRS) of harvested seeds, prediction is challenged by the high dimensionality and strong intercorrelation of NIRS data, which means that only a subset of wavelengths is informative. This study evaluates spectral variable selection models for predicting key morpho‐agronomic and quality traits in malt barley ( Hordeum vulgare L.) and assesses their performance under realistic breeding scenarios. Four NIRS‐based regularized regression models (Lasso, Enet, Ridge and a heritability‐filtered Ridge model) were tested to predict 10 morphological, agronomic and quality traits measured in two malt barley trials conducted during the 2022 and 2024 cropping seasons at three locations in Ethiopia using 100 genotypes in each trial. Model performance was evaluated across four practical breeding scenarios: within‐location unseen genotype prediction (WL‐uG), leave‐one‐location‐out prediction (LOLO), target environment unseen genotype prediction (TargetEnv) and across‐location wide adaptability (RuG). Accordingly, Cross‐validation identified stable, informative spectral predictors for each trait, scenario and trial. Among the models tested, Lasso and Enet consistently outperformed Ridge regression, with Enet showing the best predictive performance across scenarios. Prediction ability (r) ranged from 0.15 to 0.85 for quality traits, 0.16 to 0.79 for agronomic traits and 0.07 to 0.89 for morphological traits across scenarios. Thus, these findings underscore the importance of spectral predictor selection in improving predictive ability and demonstrate the transferability of PS in barley breeding.

Plant phenotyping relevance

NIRSを用いた植物形質予測とスペクトル変数選択モデルの比較・検証が研究の中心であり、複数の形態・農業・品質形質に対する予測性能を交差検証している。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.