This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility: The rapid increase in the availability of genomic data has transformed plant breeding by enabling the prediction of complex traits directly from DNA markers. Genomic prediction provides a means to accelerate selection decisions, reduce phenotyping costs and shorten breeding cycles.
Despite its promise, the practical implementation of genomic prediction remains challenging because prediction accuracy varies across datasets, populations and traits. Researchers must often evaluate multiple statistical and machine-learning approaches to identify the most suitable model, requiring substantial expertise and computational effort. To minimize the complexity and streamline this process, we have developed GenoBridge, an open-source analytical pipeline that integrates genomic prediction and genome-wide association analysis within a unified framework.
By automating model selection, performance evaluation and trait-specific genetic discovery, GenoBridge provides a scalable, reproducible and user-friendly solution to decipher biological insights from genomic datasets. Our paper is published in the journal Plant Methods. The software is available on GitHub.
Most crop traits, including grain quality, flowering time and yield, are influenced by multiple genetic variants and environmental conditions. Machine-learning models can identify patterns between DNA markers and measured traits, but their performance depends on several factors. Sample size is a major factor, and the performance of a model varies with the number of accessions.
A model that performs excellently with thousands of accessions may not work equally well with a few hundred accessions. The genetic architecture of the trait also plays a significant role. Some traits are influenced by a few genes with large effects, while others involve many genetic variants with individually small effects.
Researchers often compare multiple models and adjust their parameters through trial and error. This process becomes particularly demanding when analyzing several traits across different datasets. The pipeline uses sample size to determine analytical settings and evaluates several prediction methods, including ridge regression, random forests, gradient boosting, neural networks and ensemble models.
Extract — continue reading at the source.
