Software

G2P Datasets: A Repository of Genomic Datasets for Predictive Modeling in Plants and Animals

---

Abstract

Advances in genomic technologies have driven an exponential growth in genomic and phenotypic data, enabling increasingly sophisticated predictive models for plant and animal breeding. Large datasets spanning traits, environments, populations, and species offer tremendous potential to improve our understanding of complex traits and accelerate genetic gains. Thousands of quantitative genetics studies are published each year, and journals and funding agencies now commonly require that the associated data be made publicly available, often following FAIR (Findable, Accessible, Interoperable, and Reusable) principles. However, despite the abundance of publicly released genomic datasets, their reuse is frequently hindered by inconsistent formats and their dispersion across journals and supplementary files included in peer-reviewed articles.To improve data accessibility, reproducibility, and collaboration, we developed G2P Datasets, an open-access repository unifying and streamlining access to genomic datasets for predictive modeling in plants and animals. This resource addresses common challenges such as dispersed data, inconsistent formats, and limited reproducibility, thereby supporting genome-to-phenotype research and the benchmarking of genomic prediction models.To populate the repository, we curated more than 100 genomic datasets from diverse publications, focusing on agriculturally and veterinary relevant species. Eligible datasets had to include DNA genotypes linked to phenotypes and be freely downloadable without registration. For each dataset, we compiled detailed metadata, including sample size, trait descriptions, and citation information. Rather than hosting raw data, the repository provides R scripts that retrieve, format, and analyze datasets directly from their sources, ensuring both data integrity and scalability. The G2P Datasets repository currently includes more than 100 data sets covering more than 60 plant and animal species. For 42 of the data sets, we also provide formatted R-data objects that can be downloaded from the G2P Dataset GitHub repository or accessed through Kaggle. A Shiny web application allows users to browse, filter, and submit new datasets.To illustrate the utility of G2P Dataset, we performed a benchmark analysis using 42 datasets to compare the cross-validation prediction performance of four genomic prediction models: GBLUP, BayesB, XGBT, and Reproducing Kernel Hilbert Spaces (RKHS) Regressions using kernel averaging (RKHS-KA). In general, prediction accuracy increased linearly with trait heritability. The differences in prediction accuracy between models were typically small; however, the RKHS-KA model outperformed others in 60% of trait-dataset combinations, and the linear models (GBLUP and BayesB) consistently outperformed XGBT in ~90% of the cases.G2P Datasets provides a scalable, user-friendly platform for accessing, benchmarking, and contributing genomic datasets relevant to plant and animal breeding. By consolidating metadata and providing standardized scripts for data access and analysis, the repository enhances data findability, accessibility, and reproducibility in accordance with FAIR principles.

Keywords: 2026

How to Cite:

Aguate, A., Watson, M., Neelam, R., Deng, G., Dekkers, J. C., Steibel, J., Cheng, H. & de los Campos, G., (2026) “G2P Datasets: A Repository of Genomic Datasets for Predictive Modeling in Plants and Animals”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286866. doi: https://doi.org/10.31274/wcgalp.24148

Rights: 1