With the continuous growth of the global population, insufficient food production has become an urgent problem to be solved in most countries. At present, using artificial intelligence technology to improve suitability between land and crop varieties to increase crop yields has become a consensus among agricultural researchers. However, there are still many problems in existing works, such as limited crop phenotypic data and the poor performance of artificial intelligence models. In this regard, we take maize as an example to collect a large amount of environmental climate and crop phenotypic traits data at multiple experimental sites and construct an extensive dataset. Then, we introduce a graph neural network model to learn crop suitability evaluation and finally achieve a good evaluation effect. The evaluation results of the model can not only provide a reference for expert evaluation but also judge the suitability of the variety to other test trial sites according to the data of the current one, so as to guide future breeding experiments.

Genomic selection plays a crucial role in breeding programs designed to improve quantitative traits, particularly considering the limitations of traditional methods in terms of accuracy and efficiency. Through the integration of genomic data, breeders are able to obtain more accurate predictions of breeding values. In this study, we proposed and evaluated four deep learning architectures—CNN-LSTM, CNN-ResNet, LSTM-ResNet, and CNN-ResNet-LSTM—that are specifically designed for genomic prediction in crops. After conducting a comprehensive evaluation across multiple datasets, including those for wheat, corn, and rice, the LSTM-ResNet model exhibited superior performance by achieving the highest prediction accuracy in 10 out of 18 traits across four datasets. Additionally, the CNN-ResNet-LSTM model demonstrated notable results, showcasing the best predictive performance for four traits. These findings underscore the efficacy of hybrid models in identifying complex patterns, as they integrate skip connections to mitigate the vanishing gradient problem and enable the extraction of hierarchical features while elucidating intricate relationships among genetic markers. Our analysis of SNP sampling indicated that maintaining SNP counts within the range of 1000 to the full set significantly influences prediction efficiency. Furthermore, we conducted a comprehensive comparative analysis of predictive performance among random selection, marker-assisted selection, and genomic selection utilizing wheat datasets. Collectively, these results provide significant insights into crop genetics, enhancing breeding predictions and advancing global food security and sustainability.

Many important plants’ agronomic traits, such as crop yield, stress tolerance, and other traits, are controlled by multiple genes and exhibit complex inheritance patterns. Traditional breeding methods often encounter difficulties in dealing with these traits due to their complexity. However, genomic selection (GS), which utilizes high-density molecular markers across the entire genome to facilitate selection in breeding programs, excels in capturing the genetic variation associated with these traits. This enables more accurate and efficient selection in breeding. The traditional crop genome selection model, based on statistical methods or machine learning models, often treats samples as independent entities while neglecting the abundance latent relational information among them. Consequently, this limitation hampers their predictive performance. In this study, we proposed a novel crop genome selection model based on hypergraph attention networks for genomic prediction (HGATGS). This model incorporates dynamic hyperedges that are designed based on sample similarity to validate the efficacy of high-order relationships between samples for phenotypic prediction. By introducing an attention mechanism, it assigns weights to different hyperedges and nodes, thereby enhancing the ability to capture kinship relationships among samples. Additionally, residual connections are incorporated between hypergraph convolutional layers to further improve model stability and performance. The model was validated on datasets for multiple crops, including wheat, corn, and rice. The results showed that HGATGS significantly outperformed traditional statistical methods and machine learning models on the Wheat 599, Rice 299, and G2F 2017 datasets. On Wheat 599, HGATGS achieved a correlation coefficient of 0.54, a 14.9% improvement over methods like R-BLUP and BayesA (0.47). On Rice 299, HGATGS reached 0.45, a 66.7% increase compared to other models like R-BLUP and SVR (0.27). On G2F 2017, HGATGS attained 0.88, slightly surpassing other models like R-BLUP and BayesA (0.87). We conducted ablation experiments to compare the model’s performance across three datasets, and found that the model integrating hypergraph attention and residual connections performed optimally. Subsequent comparisons of the model’s prediction performance with dynamically selected different k values revealed optimal performance when K = (3,4). The model’s prediction performance was also compared across different single nucleotide polymorphisms (SNPs) and sample sizes in various datasets, with HGATGS consistently outperforming the comparison models. Finally, visualizations of the constructed hypergraph structures showed that certain nodes have high connection densities with hyperedges. These nodes often represent varieties or genotypes with significant impacts on traits. During feature aggregation, these high-connectivity nodes contribute significantly to the prediction results and demonstrate better prediction performance across multiple traits in multiple crops. This demonstrates that the method of constructing hypergraphs through correlation relationships for prediction is highly effective.

The phenotype (P) of a crop is determined by the genotype (G), environment (E), and genotype-by-environment (G × E) interaction, expressed as P = G + E + G × E. Thus, studying G × E interactions is essential for phenotypic research. Traditional methods of crop phenotypes and adaptability based on G × E interaction analysis, based on large ecological regions, fail to account for year-to-year environmental changes and the blurring of region boundaries, leading to inaccurate insights into the relationship between genotypes and environmental factors. To address these issues, this study divided the research area into small ecological regions through the clustering of meteorological data, providing a more accurate framework for studying G × E interactions in maize. To ascertain the optimal method for ecological region delineation, the yield variance (SYV), the Davies–Bouldin Index (DBI), and the Silhouette Index (SI) were used to evaluate and compare the performance of the K-Means, Autoencoder K-Means (Ae-KM), and Deep K-Means Clustering Neural Network (DKMCNN) methodologies. The DKMCNN surpassed other methodologies and was selected for delineation. Based on this delineation result, the interactions between genotypes and the environment on maize were investigated and clarified using genome-wide association analysis (GWAS) and analysis of variance (ANOVA). Ultimately, through the analysis of maize field trial data from 2020 to 2021, we identified up to 108 single-nucleotide polymorphisms (SNPs) in 2020 and 153 SNPs in 2021 that exerted significant effects on maize yield and exhibited strong correlations with environmental factors, including temperature, cumulative precipitation, and cumulative sunshine duration.

The integration of genotype with envirotype is essential for achieving precise phenotypic prediction. However, there is currently a scarcity of datasets derived from variety approval testing trials that encompass a broad spectrum of trial locations, conform to standardized management protocols, and which incorporate comprehensive records of environmental variables. This study introduces MaizeGEP, a dataset consisting of 260 hybrid maize varieties from national maize variety regional trials. This dataset includes 12,233 selected tag single nucleotide polymorphisms for each variety, phenotypic survey data on 11 traits across 2382 year-county locations, and daily meteorological records. We utilized this dataset to conduct analyses on the clustering of 2382 year-county locations, the population structure of the 260 varieties, and genome-wide association studies. Furthermore, a novel mixture of experts (MoE) framework incorporating genotype—envirotype to phenotype (GE2P) algorithms was employed for best linear unbiased estimator (BLUE) value prediction and phenotypic prediction. Additionally, several machine learning and deep learning algorithms, including Bayesian methods, support vector machines (SVM), LightGBM, multilayer perceptron (MLP), DeepGS, DEM, and Cropformer, were utilized to validate the effectiveness of the dataset and improve phenotypic prediction accuracy. The findings suggest that the MaizeGEP dataset serves as a valuable resource for investigating the relationship among genotype, envirotype, and phenotype, as well as predicting cross-environmental performance. This underscores the importance of encouraging researchers to utilize this dataset in developing sophisticated GE2P models. Such models can aid plant breeders in selecting new varieties and facilitating their deployment across diverse regions. MaizeGEP is publicly accessible at http://user.ebreed.cn:9992/scb/.

Enhancing the precision of crop variety adaptability evaluation is critical for maximizing yields and ensuring food security under climate change. Traditional methods relying on expert knowledge or process-based models face scalability challenges and difficulties in capturing complex variety–environment interactions. This study proposes a scalable graph-based collaborative filtering framework for assessing suitable planting locations for hybrid maize varieties, reformulating adaptability evaluation as a ranking task. The framework integrates two key innovations: a multi-layer perceptron (MLP) encoding module that leverages 15 variety trait features and 24 environmental variables to address the cold-start problem, enabling recommendations for previously unseen varieties; and a BPR+ loss function that distinguishes unsuitable environments from merely unplanted ones to accommodate the unique sparsity structure of agricultural trial data. Evaluated against six state-of-the-art graph-based recommendation models on a dataset of 189 locations and 1102 hybrid maize varieties across China’s major agro-ecological zones, our method consistently outperforms existing approaches, achieving ∼5% improvement over the strongest baseline. The MLP module attains cold-start NDCG above 0.8, and BPR+ yields significant gains in Recall@1–4 and NDCG@1–5. A case study of the widely cultivated variety ZD958, validated against both national promotion data (2019–24) and multi-year field-trial yields from 40 locations (2017–24), confirms that the model’s rank-ordered recommendations align closely with independent yield evidence: top-ranked locations achieve both the highest productivity and the greatest yield stability. This framework provides an AI-based decision-support tool for variety placement optimization and climate-resilient agriculture.