Genomic Prediction Powered by Multi-Omics Data

OSVAL A. MONTESINOS-LOPEZ^1*Abelardo Montesinos-López²Alejandro Brandon³Ivan Delgado-Enciso¹Moises Chavira-Flores⁴Jose Crossa⁵Flavio Breseghelo⁵Susanne Dreisigacker⁵Jin Sun⁶Guillermo Sebastián Gerard⁵Paolo Vitale⁵Rodomiro Ortiz^7*

• ¹Universidad de Colima, Colima, Mexico
• ²Universidad de Guadalajara, Guadalajara, Mexico
• ³Institut National des Sciences Appliquees de Lyon, Villeurbanne, France
• ⁴Universidad Nacional Autonoma de Mexico, Mexico City, Mexico
• ⁵Centro Internacional de Mejoramiento de Maiz y Trigo, Texcoco, Mexico
• ⁶Yanshan University, Qinhuangdao, China
• ⁷Swedish University of Agricultural Sciences, Uppsala, Sweden

Genomic selection (GS) has changed plant breeding by enabling early and accurate prediction of complex traits. However, its predictive performance is often constrained by the limited information captured through genomic markers alone, especially for traits influenced by intricate biological pathways. To address this, the integration of complementary omics layers-such as transcriptomics and metabolomics-has emerged as a promising strategy to enhance prediction accuracy by providing a more comprehensive view of the molecular mechanisms underlying phenotypic variation. We used three data sets each collected under a single-environment condition, which allowed us to isolate the effects of omics integration without the confounding influence of genotype-by-environment interaction. We assessed 24 integration strategies combining three omics layers: genomics, transcriptomics, and metabolomics. These strategies encompassed both early data fusion (concatenation) and model-based integration techniques capable of capturing non-additive, nonlinear, and hierarchical interactions across omics layers. The evaluation was conducted using three real-world datasets from maize and rice, which varied in population size, trait complexity, and omics dimensionality.Our results indicate that specific integration methods-particularly those leveraging model-based fusion-consistently improve predictive accuracy over genomic-only models, especially for complex traits. Conversely, several commonly used concatenation approaches did not yield consistent benefits and, in some cases, underperformed. These findings underscore the importance of selecting appropriate integration strategies and suggest that more sophisticated modeling frameworks are necessary to fully exploit the potential of multi-omics data. Overall, this work highlights the value and limitations of multi-omics integration for genomic prediction and offers practical insights into the design of omics-informed selection strategies for accelerating genetic gain in plant breeding programs.