Multi-trait random regression models increase genomic prediction accuracy for a temporal physiological trait derived...

Area of Science:

• Plant breeding and genetics
• Genomics
• High-throughput phenotyping

Background:

• Random regression models (RRM) are established for genomic inference in animals but less explored in plants.
• High-throughput phenotyping (HTP) enables extensive data collection for temporal traits.
• Integrating multiple temporal traits for genomic prediction in plants requires advanced statistical frameworks.

Purpose of the Study:

• To demonstrate the utility of multi-trait RRM (MT-RRM) for genomic prediction of daily water usage (WU) in rice.
• To jointly model WU with shoot biomass (projected shoot area, PSA) using MT-RRM.
• To assess the impact of PSA data availability on WU prediction accuracy.

Main Methods:

• Utilized MT-RRM with quadratic Legendre polynomials to model additive genetic and permanent environmental effects for WU and PSA in 357 rice accessions.
• Phenotyped WU and PSA daily over 20 days using HTP.
• Evaluated predictive abilities using two cross-validation scenarios, with and without PSA data in the testing population.

Main Results:

• Observed weak to strong genomic correlations between WU and PSA across imaging days (0.29-0.87 and 0.38-0.80).
• MT-RRMs consistently outperformed single-trait RRM in predictive ability for WU.
• Prediction accuracies for WU significantly improved when PSA records were incorporated into the MT-RRM framework.

Conclusions:

• MT-RRM provides an effective framework for genomic prediction of temporal physiological traits in plants.
• Joint modeling of WU and PSA enhances prediction accuracy, offering a valuable tool for plant breeding.
• This approach facilitates the prediction of traits that are challenging to measure in large populations.