In crops, important agronomic traits such as yield, disease resistance, stress tolerance and quality, are fundamentally shaped by genetic variation. Rapid advances in high-throughput sequencing have generated vast amounts of crop genomic data, creating new opportunities for deep learning to uncover the genetic basis of these traits. However, genomic resources remain unevenly distributed across crop species, and models developed for one species often transfer poorly to another, limiting the broader use of accumulated genomic knowledge.
In a study published in The Crop Journal , a research team led by Professor Zhenyu Yue from Anhui Agricultural University has developed Crop-GPA 2.0, a novel deep-learning framework for transferable SNP-level genotype–phenotype association prediction across crop species. By combining hierarchical genomic representations with cross-species pre-training and trait-aware learning, Crop-GPA 2.0 enables effective transfer of SNP–trait association knowledge among crops, providing a scalable approach for decoding trait-associated genetic variation and prioritizing candidate functional SNPs.
“Crop-GPA 2.0 builds on our continuing efforts to decode genotype–phenotype associations in crops,” says Yue. “While Crop-GPA 1.0 focused on gene-level associations, Crop-GPA 2.0 moves to the finer resolution of individual SNPs and extends association modeling across species and traits, bringing us closer to identifying functional variants underlying important agronomic traits.”
At the core of Crop-GPA 2.0 is a hierarchical genomic representation strategy that integrates multi-scale DNA sequence and structural features. “By capturing complementary information around candidate SNPs, this strategy provides a richer genomic representation for distinguishing variants associated with important crop traits,” adds Yue.
Building on these representations, cross-species pre-training and trait-aware learning allow Crop-GPA 2.0 to capture genomic patterns shared among crops while retaining signals relevant to specific traits. This combination enables knowledge learned across diverse genomic backgrounds to be reused for new crop–trait prediction tasks rather than treating each species independently.
“The key is to capture what can be shared across crops without losing trait-specific information,” explains Yujia Gao, the study’s first author. “By bringing these two levels of information together, Crop-GPA 2.0 provides a more transferable way to decode genotype–phenotype relationships across diverse crops.”
The researchers evaluated Crop-GPA 2.0 across representative crops, including rice, maize and wheat. The framework outperformed existing methods across multiple prediction tasks and retained robust performance during cross-species transfer and when training data were reduced, demonstrating strong generalization across diverse genomic backgrounds.
Beyond predictive performance, the researchers assessed whether Crop-GPA 2.0 could recover biologically meaningful genetic signals. High-confidence SNP predictions were supported by previously reported quantitative trait locus (QTL) evidence, indicating that the framework can prioritize trait-relevant variants with independent biological support.
“These findings show that Crop-GPA 2.0 goes beyond improving computational prediction,” says Gao. “It provides biologically meaningful SNP-level evidence that can help bridge genotype–phenotype association modeling and functional variant discovery.”
To make the framework broadly accessible, the team also developed an online Crop-GPA 2.0 platform, integrating genotype–phenotype association data with species- and trait-specific SNP prediction, visualization and online analysis.
Crop-GPA 2.0 provides a scalable AI framework for making better use of genomic knowledge across crop species. The approach opens new opportunities for functional variant discovery, crop functional genomics and precision breeding.
###
Contact Author:
Zhenyu Yue
E-mail address: zhenyuyue@ahau.edu.cn
The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 200 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).
The Crop Journal
Computational simulation/modeling
Not applicable
Monitoring of overwintering leaf age by integrating phenological, temporal, and thermal data: An indicator for assessing winter wheat seedling condition
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.