Improving prediction accuracy of genomic-selection by using multiple-trait models

• Applicant: Dr. Tigist Mideksa Damesa
• Subject Area: Plant Breeding and Plant Pathology
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 469205605

The production of food will have to increase continuously in order to feed the rapidly growing world population. Plant breeding programs are contributing to address the challenges of increasing the production of foods. Genomic Selection (GS) is an effective breeding tool that provides an opportunity to increase food production at a faster rate, by reducing selection cycle times and phenotyping costs. GS uses phenotypic and genotypic data for model calibration and then estimates genomic estimated breeding values (GEBV) for non-phenotyped individuals. Even though GS is a powerful tool, its reliability and success depend on its predictive accuracy, which in turn is affected by many factors. One approach to improve prediction accuracy of GS is by using multi-trait (MT) GS. MTGS allows obtaining a more accurate evaluation by exploiting genetic correlation between traits. Studies have shown that MTGS using secondary traits and/or high-throughput data which are correlated to traits of interest can increase predictive accuracy of GS, and thereby increase genetic gain. Moreover, MTGS is useful to improve several desirable traits simultaneously. Therefore considering MTGS is very promising approach.The predictive accuracy of GS can also be significantly affected by the statistical method used to estimate marker effects. Particularly for GS in plant breeding a statistical model that accounts several sources of variations under study is required. The phenotypic data for GS is usually based on multi-environment trials (MET), and its analysis is complex, because the statistical method must account for replication effects, within trial spatial and heterogeneous error variations, marker-by-environment interaction (MEI), genotype-by-environment interaction (GEI), and non-additive genetic effects. Accommodating all these factors promised to increase prediction accuracy of GS. This is well established for single trait GS. However, the application of these methods in MTGS is lacking behind. Therefore the main objective of this study is to advance prediction accuracy of GS in an MT context, using phenotypic data from MET and advanced statistical approaches that account for spatial adjustment, GEI, and MEI, implementing remedies for within and between variance heterogeneity, and modeling of both additive and non-additive effects. Towards the end of the project, we will develop a joint modeling framework and pipeline to integrate the different parts above.

• DFG Programme: Research Grants