JOINT GENETIC EVALUATION OF FIELD AND STA TION TEST IN THE HUNGARIAN LARGE WHITE AND LANDRACE POPULATIONS

Variance and covariance components were estimated for the Hungarian La ndrace (LR) and Large White (LW) populations using an individual anima l model with a derivative free and - for the first time - a derivative based REML procedure for traits that are used in the selection index. A complete data set from the National Hungarian genetic swine evaluat ion program from the field and station test were analyzed separately a nd jointly, the latter resulting in around 250 000 mixed model equatio ns with 55 (co)variance components to be estimated simultaneously in a six trait model. This was only made possible by using a quasi-Newton procedure (QN) with optimization on the Cholesky factor of the covaria nce matrices which converged after 48 days of CPU time. Using QN based on first derivatives or analytical gradients reduced the computing ti me to 41 hours facilitating for the first time the joint multivariate evaluation of real life data sets used in genetic evaluation. Besides computing the gradients much faster by the exact procedure also a fast er convergence could be observed reducing the number of iterations to around a third. The total speedup was in the order of 50. Apart from t he traditional ratio traits like daily gain a genetic analysis was als o performed for the directly measured main components like days on tes t. Heritabilies tended to be somewhat higher for direct traits. Estima tes for days on test, feed intake, valuable cuts, meat quality and age and back fat in the field were .39, .51, .62, .13, .20, .10 for LR an d .31, .46, .66, .08, .24, and .19 for LW,respectively. Genetic trends were estimated as the regression of BLUPs on year of test. The estima tes for days on test, feed intake, valuable cuts, meat quality and age and back fat were -1.1 days, -2.19 kg, 115g, -.002 points, -.6 days a nd -.008 mm for LR, and -.55 days, -.678 kg, 53g, -.007 points, -1.0 d ay and -.02 mm for LW.