Achieving adequate levels of genetic gain for yield and all target traits is increasingly difficult, but effective strategies for implementing the new technologies of genomic selection (GS) and high-throughput phenotyping (HTP) can help predict and identify the best lines and parent combinations. The Delivering Genetic Gain in Wheat approach will: 1) develop and implement a practical prediction-based selection strategy using remote sensing, genomic, and pedigree data in order to accurately select for yield prior to yield testing where seed is limited; 2) develop and evaluate methodologies for selecting the best parent combinations from prediction data. The prediction-based selection strategy developed and implemented in the CIMMYT spring wheat breeding program will provide proof of concept and is expected to deliver greater rates of genetic gain.
The GS and HTP objective will initiate a genomic selection and high-throughput phenotyping prediction methodology, integrating remote sensing, genomic, and pedigree data, that will be developed specifically for breeding candidates prior to yield testing where selection accuracies for yield are currently low due to smaller field plots and limited seed availability. Developing methodologies for HTP on small plot sizes and with no replication of breeding candidates will be critical. Methodologies developed will be validated using replicated yield trials data to identify a suitable HTP methodology for these early stage breeding candidates. One of the novel proposed methodologies, the Wheat Walker, is a new low-cost field mobile robot platform being developed by Dr. Joshua Peschel at the University of Illinois at Urbana-Champaign and field implemented and assessed on wheat by Dr. Michael Gore and Ph.D. Student Ms. Margaret Krause at Cornell University. Specially, the Wheat Walker is a small tactical and semi-autonomous, ground-based robotic platform with sensors capable of mapping the physical structure of a cereal canopy in situ and measuring its local climatic parameters.
Using the same set of data from the small plot HTP experiment, prediction models incorporating HTP and pedigree data will be compared with models integrating HTP and genome-wide marker data. These models will also be compared to conventional genomic and pedigree prediction models that do not incorporate HTP data. For each combination of HTP method and prediction model, prediction accuracies and cost per line will be considered to determine the expected rate of genetic gain from each scenario given a fixed budget. The methodology that maximizes genetic gain will be identified. The best selection methodology in terms of expected genetic gain assuming a fixed budget will be implemented at the F4:F5 stage. For each trait that is part of the breeding goal, general functions relating the trait value with the net merit will be identified. Using these functions, methods for optimizing cross combinations will be evaluated using simulation. The best methods will be identified for further testing.