Scholarlink

Key Message <p>An explainable Genomic Selection framework was proposed to identify significant SNPs of wheat yield traits via integrating hybrid feature selection, fusing multi-stage heterogeneous model ensembles, and implementing SHAP-driven interpretable prediction.</p> Abstract <p>Improving wheat yield is a core goal for breeders in molecular breeding. Genomic Selection (GS) is crucial to this goal. However, wheat yield-related GS studies have long faced challenges of high-dimensional genotypic data and insufficient biological interpretability. This study improved the cutting-edge MFMGP algorithm and proposed a multi-stage hybrid feature fusion interpretable GS framework, named TRXB, to enhance the prediction accuracy of wheat yield traits’ Genomic Estimated Breeding Values (GEBVs). TRXB first built high-information-density SNP subsets via triple screening strategies, including GWAS significance, Meta-QTL pathways, and Knowledge-Guided Module (KGM) interactions, then utilized the Optuna technique to select 4 high-performance algorithms from 14 mainstream GS methods covering linear models, machine learning, and deep learning, and finally utilized the SHapley Additive exPlanations (SHAP) to analyze wheat yield contributions and interactions of significant SNPs. Results show that on the ZSS dataset, TRXB achieved a Thousand-Grain Weight (TGW) prediction Pearson Correlation Coefficient (PCC) of 0.862, 4.86% above 14 mainstream GS models on average, 3.48% above the benchmark MFMGP, and 12.3% above the best mainstream GS for the other three traits. On the external ZJ dataset, its PCC for TGW and Grain Number per Spike (GN) rises by 7.35% and 16.23% respectively, demonstrating strong generalization. By combining high accuracy, stability, and interpretability, TRXB offers a new paradigm for smart crop breeding.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TRXB: a multi-stage hybrid feature fusion interpretable genomic selection ensemble model for improving the prediction accuracy of wheat yield traits

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月19日

查看全文
  • Haoqin Ren,
  • Jun Zheng,
  • Limin Wang,
  • Xianggeng Huang,
  • Shanghui Zhang,
  • Shijie Qiu,
  • Ran Qin,
  • Chunhua Zhao,
  • Yongzhen Wu,
  • Han Sun,
  • Guangchen Liu,
  • Mei Song,
  • Shusong Zheng,
  • Fa Cui

摘要

Key Message

An explainable Genomic Selection framework was proposed to identify significant SNPs of wheat yield traits via integrating hybrid feature selection, fusing multi-stage heterogeneous model ensembles, and implementing SHAP-driven interpretable prediction.

Abstract

Improving wheat yield is a core goal for breeders in molecular breeding. Genomic Selection (GS) is crucial to this goal. However, wheat yield-related GS studies have long faced challenges of high-dimensional genotypic data and insufficient biological interpretability. This study improved the cutting-edge MFMGP algorithm and proposed a multi-stage hybrid feature fusion interpretable GS framework, named TRXB, to enhance the prediction accuracy of wheat yield traits’ Genomic Estimated Breeding Values (GEBVs). TRXB first built high-information-density SNP subsets via triple screening strategies, including GWAS significance, Meta-QTL pathways, and Knowledge-Guided Module (KGM) interactions, then utilized the Optuna technique to select 4 high-performance algorithms from 14 mainstream GS methods covering linear models, machine learning, and deep learning, and finally utilized the SHapley Additive exPlanations (SHAP) to analyze wheat yield contributions and interactions of significant SNPs. Results show that on the ZSS dataset, TRXB achieved a Thousand-Grain Weight (TGW) prediction Pearson Correlation Coefficient (PCC) of 0.862, 4.86% above 14 mainstream GS models on average, 3.48% above the benchmark MFMGP, and 12.3% above the best mainstream GS for the other three traits. On the external ZJ dataset, its PCC for TGW and Grain Number per Spike (GN) rises by 7.35% and 16.23% respectively, demonstrating strong generalization. By combining high accuracy, stability, and interpretability, TRXB offers a new paradigm for smart crop breeding.