DEVELOPMENT OF STATISTICAL METHODS FOR THE ANALYSIS OF GENETIC (CO)VARIANCE MATRICES CALCULATED USING SINGLE NUCLEOTIDE POLYMORPHISMS
Animal Improvement Programs
2012 Annual Report
1a.Objectives (from AD-416):
ARS Project Plan 1265-31000-096-00D has two objectives that directly relate to this agreement. The first is to develop methods to incorporate high-density genomic data in predictions of genetic merit, and the second is to investigate correlations among traits to efficiently combine evaluations to select for healthy dairy animals capable of producing quality milk at a low cost in many environments. The Cooperator has the expertise and infrastructure to develop statistical methods for the analysis of genome- and chromosome-wide (co)variance matrices calculated from genetic marker effects. Such methods may be useful for understanding the biology of economically important traits in dairy cattle, as well as identifying important nodes in gene networks.
1b.Approach (from AD-416):
Single nucleotide polymorphism (SNP) effects estimated independently in the Italian and U.S. Brown Swiss cattle populations by the Cooperator and ARS, respectively, will be used to develop methods for comparing genetic (co)variance matrices for various traits of economic importance in those breeds. Methods developed will be reported in the scientific literature. ARS and the Cooperator will jointly develop the statistical methods using a shared set of simulated data. The methods will be applied by AIPL to U.S. data and by the Cooperator to Italian data. Results will be used to compare the two populations.
The project is related to in-house objectives 2 (characterize phenotypic measures of dairy practices and provide the industry with information for determining impact of herd management decisions on profitability) and 3 (improve accuracy of prediction of economically important traits currently evaluated). An approach has been developed to compare statistically the structure of genomic (co)variance matrices to determine when two (or more) such matrices differ from one another. Eigenvalues from the decomposition of the genomic matrices by multivariate factor analysis may be compared using simple linear regression models. Two matrices with equivalent structures are expected to have eigenvalues that when regressed on one another produce an intercept of zero and a slope of one. This concept was demonstrated to hold empirically using genomic correlation matrices from bovine chromosomes 1 and 18, which have very different correlation structures, as well as for chromosomes 1 and 2, which have very similar structures. However, several questions remain to be answered, including how best to account for multiple comparisons when a large number of matrices are compared with one another. Loadings of traits on factors also were shown to differ with varying correlation structures. A manuscript that describes the new methodology is being prepared for submission to a peer-reviewed journal.