1


2

 Measure of amount of information in an evaluation
 Squared correlation between EBV and true breeding value
 May be expressed as daughter equivalents (DE = k*Rel/(1Rel), where k =
variance ratio)
 Diagonal element of inverse of coefficient matrix
 Approximation used when computing inverse is too time consuming

3

 Time to compute inverse increases cubically with number of genotyped
animals
 When inversion discontinued in 2009, all animals given same contribution
from genomics
 Genomic relationship (G) with predictor animals measures information
contributed
 REL of predictor animal evaluation used for weighting

4

 Convert traditional reliability to DE
 Remove contribution from parents
 Convert back to reliability for weighting genomic relationship
 Sum weighted relationships across predictor animals (^{∑}G_{w}).
Similar to genomic future inbreeding

5

 August 2009 reliabilities from inverse converted to DE and genomic
contribution (DE_{g}) extracted
 DE_{g} predicted from ^{∑}G_{w }for 3 Holstein
groups
 8,353 bulls with traditional evaluations
 3,559 cows with traditional evaluations
 16,135 animals without traditional evaluations

6

 Straight line fit well except for low ^{∑}G_{w}
 Slopes and floor differ by breed
 Applicable across traits because same k used to calculate DE_{g}

7


8


9


10


11

 Computing time increases by (number of SNP)*(number of genotyped
animals)^{2}
 Options for calculating genomic relationships
 Get faster computer
 Run program in parallel
 Compute relationships before evaluations
 Calculate only for new animals; store old ones
 Calculate based only on SNP from 3K chip
 Use Eigenvalues & Eigenvectors to reduce processing
 If G used to calculate EBV greater precision needed

12

 Problems – greater with imputation
 Missing genotypes
 Genotype incorrect
 Remedies
 Discount for missing SNP genotypes
 Consider if parents were genotyped
 Discount uniformly for 3K chip

13

 Chip expected to ship this week
 2,882 usable SNP, plus 14 on Y Chromosome
 Unofficial evaluations expected Sept. 7
 Large initial use expected
 New genotyping laboratories

14

 Improved REL approximation implemented April 2010
 REL reduced for animals less related to predictor population
 MN controlline bulls
 Foreign bulls
 Approximation relies on G matrix
 Modification planned to account for loss in accuracy from imputation

15

