|GREDLER, BIRGIT - Collaborator|
|SEEFRIED, FRANZ - Collaborator|
|SCHULER, URS - Collaborator|
|SARGOLZAEI, MEHDI - University Of Guelph|
|HICKEY, JOHN - University Of New England|
Submitted to: Meeting Abstract
Publication Type: Proceedings
Publication Acceptance Date: 6/17/2012
Publication Date: 6/17/2012
Citation: Gredler, B., Seefried, F., Schuler, U., Van Raden, P.M., Sargolzaei, M., Hickey, J. 2012. Accuracy of genotype imputation in Swiss cattle breeds. 4th International Conference on Quantitative Genetics: Understanding Variation in Complex Traits, Programme & Book of Abstracts, p. 152, P-184. 2012.
Technical Abstract: The objective of this study was to evaluate the accuracy of imputation from Illumina Bovine3k Bead Chip (3k) and Illumina BovineLD (6k) to 54k chip information in Swiss dairy cattle breeds. Genotype data comprised of 54k SNP chip data of Original Braunvieh (OB), Brown Swiss (BS), Swiss Fleckvieh (SF), Simmental (SI), and red Holstein (HO). Genotypes of OB and BS (data set BSW) were analysed together as well as SI, SF, and HO (data set MIX). After routine genotype quality checks BSW and MIX included 3,738 animals with 39,841 validated SNP and 4,753 animals with 39,743 validated SNP, respectively. 54k genotypes of animals born between 2008 and 2011 were masked to mimic animals genotyped with the 3k and 6k chips. The 3k and 6k chip included 2,846 and 6,605 SNP for BSW and 2,715 and 6,153 SNP for MIX, respectively. Methods used for imputation were AlphaImpute, Beagle, FImpute, and Findhap V2. The accuracy of imputation was assessed by the squared correlation (R2) between true and imputed genotypes. R2 was higher for imputation from 6k than from 3k. R2 decreased with lower relationship between the 54k genotyped reference population and the 3k and 6k genotyped imputation candidates. For BSW, average R2 was highest (0.96) using AlphaImpute and FImpute when both parents of a 6k genotyped animal were genotyped for the 54k chip. R2 was between 0.85 and 0.93 when only the sire was 54k genotyped and lowest (0.79-0.92) when no direct relatives were 54k genotyped. Accuracy of imputation was highly dependent on MAF of the imputed SNP. Using FImpute R2 was between 0.67 (MAF < 0.025) and 0.96 (MAF between 0.4 and 0.5). All programs gave high imputation accuracy where FImpute slightly outperformed the other programs in terms of the R2 while AlphaImpute had lowest error rates.