Optimization and application of genome prediction model in rapeseed: flowering time, yield components, and oil content as examples
Wen-Kai Yu, Xinao Wang, Hui Wang, Wenxiang Wang, Hongtao Cheng, Desheng Mei, Lixi Jiang, Qiong Hu, Jia Liu
Abstract
Rapeseed is the second largest oilseed crop in the world with short domestication and breeding history. This study developed a batch of genomic prediction models for flowering time (FT), oil content, and yield components in rapeseed. Using worldwide 404 breeding lines, the optimal prediction model for FT and five quality and yield traits was established by comparison with efficient traditional models and machine learning (ML) models. The results indicate that quantitative trait loci (QTLs) and significant variations identified by genome-wide association study (GWAS) can significantly improve the prediction accuracy of complex traits, achieving over 90% accuracy in predicting FT and thousand grain weight. The Genomic Best Linear Unbiased Prediction (GBLUP) and Bayes-Lasso models provided the most accurate prediction overall, while ML models such as GBDT (Gradient-Boosting Decision Trees) exhibited strong predictive performance. Our study provides genome selection solution for the high prediction accuracy and selection of complex traits in rapeseed breeding. The use of a diverse panel of 404 worldwide lines ensures that the findings are broadly applicable across different rapeseed breeding programs.