Introduction

1. Results (cont.) Software & Computing (cont.) Introduction Higher density genotypes can provide markers closer to QTL, but imputation is needed for genotypes of less than highest density. Markers from multiple chips can then be combined in genomic evaluation. • Three combinations of the programs were tested: findhap run once (imputing from 3K and 50K up to HD), findhap run twice (first imputing 3K to 50K then imputing 50K to HD), and running FImpute (imputing 3K to 50K) before running findhap (imputing 50K to HD). • Several combinations of segment lengths were tested in findhap. • Imputation of 636,967 markers for 103,070 animals with findhap required 50 Gbytes of memory and 10 hours using 6 processors. • Iteration for SNP effects for 29 traits required 2 days using 6 processors. • August 2007 predictions were tested with April 2011 data Table 2. Gains in Reliability Objectives • Determine the accuracy of imputing up to 636,967 markers (HD) from 42,495 markers (50K), 2,614 markers (3K) or from 0 markers (imputed dams) using simulated data. • Determine gain in reliability from using more markers with actual data. Data • Four types of genotypes were used for this analysis: HD, 50K, 3K, and imputed dams. • The animals genotyped included 1,074 with HD, 66,540 with 50K, 33,119 with 3K, and 2,337 imputed dams. • HD genotypes were from 356 influential USA and CAN sires, 398 GBR sires, 156 other sires, 138 Beltsville research cows, and 26 other females. • To test imputation, an example simulated chromosome was used with 1% of the genotypes missing and 0.02% incorrect initially from each chip. Among all animals, 94.4% of genotypes were missing initially. Results • A maximum length of 2,000 markers and a minimum of 200 yielded the best results when findhap was run one time. • A maximum length of 1,500 markers and a minimum of 200 markers yielded the best results when findhap was run twice and when findhap and FImpute were combined . • Running FImpute and findhap yielded the best results with an average of 96.37% correctly called HD genotypes across all chip types including imputed dams (Table 1). • The average reliability gain over all traits was 0.4% (Table 2). Software & Computing Table 1. Correctly imputed genotypes. • Both findhap.f90 developed at AIPL and FImpute developed at U. Guelph and Boviteq Alliance were tested in this analysis. • The imputation rate with findhap version 2 is improved compared to version 1 results tested earlier. • Version 2 of findhap uses both long segments to improve haplotype matches for close relatives and short segments to help detect matches from more remote ancestors. Conclusions • Imputation from 50K to HD is accurate (98.9%), • The 0.4% average increase in reliability is less favorable than the 0.9% expected from simulation. • More animals with HD genotypes will improve imputation and reliability. • Multi-breed evaluation could produce larger gains than the single-breed evaluation that was investigated.