Genomic Selection with the Realized Relationship Matrix.

Nearly twenty years have passed since best linear unbiased prediction (BLUP) with an additive relationship matrix was first used to predict the performance of untested maize hybrids. Although not widely adopted in plant breeding, there is renewed interest in BLUP as a strategy for genomic selection.  Instead of using a numerator relationship matrix based on pedigrees, however, realized relationships are estimated from molecular markers.  A widely cited and important result from genomic selection theory is that BLUP with a realized relationship matrix is equivalent to estimating marker effects by ridge regression.  In the first part of the talk, this idea will be refined and shown to hold only in the limit of complete, high-density markers.  In practice, breeders are likely to use either small SNP arrays or low coverage genotyping-by-sequencing with considerable missing data.  New techniques for estimating the realized relationship matrix from such marker data will be discussed and shown to be superior to current practice.  The second part of the talk will present the initial phases of a genomic selection project to increase wheat yield, managed as a collaboration between CIMMYT and Cornell University.  The focus will be on the design and implementation of a prediction model using the realized relationship matrix.