Progress 09/01/05 to 08/31/07
Outputs
Progress Report Objectives (from AD-416) Reduce the cost and complexity of running the genetic evaluation for calving ease in U.S. dairy breeds by the threshold model; investigate possibilities of improving the approximation of accuracies of PTA. Approach (from AD-416) Computer program cblup90iod will be modified for speed. The modifications will include an option to use fixed thresholds. Part of the work will be identification of conditions that lead to slow convergence or divergence. Subsequently, appropriate modifications will ensure good convergence for a variety of models and data sets. The renumbering and pedigree analysis program renumf90 will be modified to provide data and parameter sets suitable for sire models. Having a U.S. animal pedigree file, the program will extract the sire pedigree and provide pruning, i.e., only sires relevant to the evaluation will be included. Several approximations of accuracy in threshold sire models will be compared. Methods will be based on the value of diagonals in equations based on linear and threshold models. Alternatively, accuracies will be obtained based on the marginal distribution in Bayesian method by MCMC. The comparison will be with simulated data and with a subset of the real data; MCMC is unsuitable for large data sets. Comparisons will be made for designs with different distribution of sires. As found by Rekaya et al. (2005), the threshold model as used currently may need modifications when many sires and contemporary groups have one score only. Significant Activities that Support Special Target Populations This report documents research conducted under a specific cooperative agreement between ARS and the University of Georgia (Athens, GA). Additional details of research can be found in the report for the parent project 1265-31000-085-00D, "Improving Genetic Prediction of Economic Merit of Dairy Animals". research conducted under the agreement is related to objectives 3 (identification and adaptation of statistical tools) and 4 (improved rankings for overall genetic merit). Monitoring activities for the project included meetings with the Cooperator in Raleigh, NC, Athens, GA, and San Antonio, TX, as well as e-mail and phone conversations. Several computing options were evaluated for estimating breeding values with best linear unbiased prediction properties for potentially thousands of single-nucleotide polymorphism (SNP) markers for each phenotypic record. Those options included half-stored Cholesky decomposition and Gauss-Seidel and three matrix-free strategies: Gauss- Seidel, Gauss-Seidel with residual update, and preconditioned conjugate gradients. Computing algorithms were tested on a mouse data set that included 1,928 records and 10,946 SNP markers. Based on computing times, Gauss-Seidel with residual update was determined to the method of choice for variance component estimation as well as solving for breeding values. A scientific paper on the enhancement of the PyPedal computer program to provide tools for pedigree analysis, report generation, and data visualization was prepared, submitted to Computers and Electronics in Agriculture, and published.
Impacts
(N/A)
Publications
|
Progress 10/01/05 to 09/30/06
Outputs
Progress Report 4d Progress report. This report serves to document research conducted under a specific cooperative agreement between ARS and the University of Georgia (Athens, GA). Additional details of research can be found in the report for the parent project 1265-31000-085-00D "Improving Genetic Prediction of Economic Merit of Dairy Animals"; research conducted under the agreement is related to objectives 3 (identification and adaptation of statistical tools) and 4 (improved rankings for overall genetic merit) of the parent project. Because the computer program CBLUP90IOD appeared to have a low convergence rate for threshold models, different constraints on threshold solutions were examined. The original version set the last threshold to 0. Two new options were investigated: 1) set the first threshold to 0 or 2) no constrains. The last option had an excellent convergence rate (as good as or better than thresholds set to fixed values). The change has been
incorporated into CBLUP90IOD. A computer program (PyPedal) was enhanced to provide tools for pedigree analysis, report generation, and data visualization.
Impacts
(N/A)
Publications
|
|