Predicting complex phenotypes using multi-omics data in maize

Area of Science:

• Plant genetics and breeding
• Systems biology
• Agricultural biotechnology

Background:

• Predicting complex plant traits is challenging due to interactions between genetics, regulation, and environment.
• Accurate trait prediction and identification of genetic elements are crucial for crop improvement and biotechnology.

Purpose of the Study:

• To evaluate if multi-omic datasets (genomic, transcriptomic, phenomic) improve predictive accuracy for diverse maize phenotypes across multiple environments.
• To compare the performance of linear and nonlinear models using single- and multi-omics inputs.

Main Methods:

• Utilized genomic markers, field-based transcriptomic data, and drone-derived phenomic data for 129 maize phenotypes across nine environments.
• Trained and compared linear (rrBLUP) and nonlinear (support vector regression) models with single- and multi-omics data.
• Analyzed model feature weights to understand the distribution of predictive signals.

Main Results:

• Multi-omics models consistently outperformed single-omics models in predicting maize traits.
• Transcriptomic data enabled accurate cross-environment trait prediction and captured genotype-by-environment interactions.
• Phenomic data showed lower predictive power alone but improved predictions for specific traits like root architecture.

Conclusions:

• Integrating transcriptomic and phenomic data with genotypes enhances trait prediction accuracy and model generalizability across environments.
• This multi-omics approach provides deeper insights into the genetic and regulatory architecture of agriculturally important maize traits.
• Complex traits arise from coordinated, network-level genetic processes rather than a few dominant loci.