Accurately predicting complex agronomic traits remains a major hurdle in crop breeding. This study shows that optimized genomic prediction models can reliably forecast flowering time, yield components, and oil content in rapeseed using genome-wide genetic data. By combining trait-linked genetic variants with both conventional statistical methods and machine-learning approaches, the researchers achieved high prediction accuracy across several economically important traits. Incorporating genome-wide association (GWAS) signals substantially improved performance, particularly for flowering time and thousand-seed weight. The findings underscore the potential of genomic prediction to speed up selection, shorten breeding cycles, and support simultaneous improvement of multiple traits in oilseed crops.
Rapeseed is among the world’s leading oil crops, but breeding progress is limited by the genetic complexity of traits such as flowering time, seed yield, and oil content. These characteristics are controlled by many genes with small effects and are strongly shaped by population structure and domestication history. As a result, traditional breeding methods and marker-assisted selection often struggle to capture the full genetic architecture efficiently. While whole-genome sequencing now provides large volumes of genetic information, translating these data into accurate trait predictions remains challenging — highlighting the need for robust, practical genomic prediction strategies.
A team from the Oil Crops Research Institute of the Chinese Academy of Agricultural Sciences, together with collaborating institutions, addressed this challenge in a study published in Horticulture Research on April 30, 2025 (DOI: 10.1093/hr/uhaf115). Using a globally diverse panel of 404 rapeseed breeding lines, the researchers developed and evaluated a comprehensive genomic prediction framework for improving key traits. Their work offers a clear roadmap for integrating genomic prediction into modern rapeseed breeding programs, according to a press release.
The study analyzed high-density resequencing data from 404 rapeseed accessions representing spring, winter, and semi-winter ecotypes from around the world. After identifying more than 23 million high-quality genetic variants, the team combined phenotypic data collected across two growing seasons with GWAS to pinpoint loci linked to flowering time, yield-related traits, and oil content. They identified 22 significant quantitative trait loci, many of which were shared across traits, suggesting overlapping genetic control.
To assess predictive performance, seven genomic prediction models—including GBLUP, Bayes–Lasso, and multiple machine-learning algorithms—were tested using different sets of genetic features. Models that incorporated GWAS-associated variants consistently outperformed those based on randomly selected or dimension-reduced markers. Prediction accuracy surpassed 90% for flowering time and thousand-seed weight, while yield and oil-related traits reached more than 80%. Traditional models such as GBLUP and Bayes–Lasso proved highly stable, especially with moderate sample sizes, while machine-learning approaches performed competitively when paired with suitable feature-selection strategies. Importantly, the study found that including both major- and minor-effect variants maximized predictive power while helping reduce genotyping costs.
The researchers conclude that genomic prediction can address one of breeding’s central challenges: improving traits controlled by many interacting genes. By selecting trait-associated variants and applying well-matched prediction models, breeders can estimate performance earlier in development—often before plants reach maturity. This accelerates breeding progress and makes it easier to improve flowering time, yield, and oil quality at the same time, even when these traits are traditionally difficult to optimize together.
The framework described in this study can be readily applied in rapeseed breeding programs worldwide. By reducing dependence on lengthy field trials, breeders can make faster and more accurate selection decisions and increase genetic gain. The approach also supports more cost-effective genotyping, improving accessibility for large-scale breeding efforts. Beyond rapeseed, the methodology provides a transferable model for other crops with complex trait architectures, advancing data-driven breeding to meet rising global demand for edible oils and sustainable agriculture.
