Restricted Maximum Likelihood Estimation of ...

Restricted maximum likelihood estimation of variance components from field data for number of pigs born alive M. T. See, J. W. Mabry and J. K. Bertrand J ANIM SCI 1993, 71:2905-2909.

The online version of this article, along with updated information and services, is located on the World Wide Web at: http://jas.fass.org/content/71/11/2905

www.asas.org

Downloaded from jas.fass.org by guest on July 13, 2011

Restricted Maximum Likelihood Estimation of Variance Components from Field Data for Number of Pigs Born Alive’12 M. T. See, J. W. Mabry, and J. K. Bertrand Animal and Dairy Science Department, The University of Georgia, Athens 30602-2771

ABSTRACT Variance components for number of pigs born alive (NBA) were estimated from sow productivity field records collected by purebred breed associations. Data sets analyzed were as follows: Hampshire ( n = 13,5371, Landrace ( n = 10,822), and Spotted ( n = 3,949). Variance components for service sire, sire of sow, dam of sow, and residual effects on NBA (adjusted for parity) were estimated. The singletrait model included relationships between service sires, sires of sows, and dams of sows. The model was implemented using an expectation maximization

(EM ) REML algorithm. A sparse-matrix solver was also used. Heritability estimates for NBA were .13, .13, and .12 for Hampshire, Spotted, and Landrace, respectively. Estimates of maternal genetic (co)variances ( m 2 ) expressed as a proportion of the phenotypic variance were .05, .01, and .03 for Hampshire, Spotted, and Landrace, respectively. Results indicated that service sires account for 1 to 2% of the total variation for NBA. Genetic effects influencing NBA seem to be small in these data sets, but selection for increased NBA should be effective.

Key Words: Pigs, REML, Variance Components, Litter Size

J. h i m . Sci. 1993. 71:2905-2909

Introduction Genetic evaluation is conducted on reproductive traits, such as number born alive ( NBA), as an aid to accelerate genetic improvement. This improvement requires knowledge of the heritability for the trait of interest. Variance component estimates are also needed for additional random effects that may be included in the genetic prediction model. These estimates are specific for a given population over a defined period of time. Misztal (1990) stated that “the accuracy of estimates of variance components is dependent on the choice of data method and model.” It is generally regarded that the best variance component estimates would come from a large data set, using a REML procedure and a full animal model. An analysis that meets all of these objectives, although ideal, can rapidly become too computer-intensive for practical use when one analyzes large field data sets. To complete this difficult task, researchers who have used large ( > 3,000 litter records) swine field data sets

‘This research was supported by State and Hatch funds allocated to the Georgia Agric. Exp. Sta. 2Data was provided by the Hampshire Swine bgistry, the American Landrace Association, and the National Spotted Swine Record. Received November 24, 1992. Accepted June 28, 1993.

have used a simpler procedure (Strang and King, 1970; Strang and Smith, 1979; McCarter et al., 19871, a simpler model (McCarter et al., 1987; Kaplon et al., 1991), or pooled results across subsets of the data set (Kaplon et al., 1991). The objective of the present study was to estimate variance components for NBA in Hampshire, Landrace, and Spotted swine field data, using a procedure that balances accuracy, time, and cost. Estimates of genetic and environmental parameters within each breed were obtained.

Materials and Methods

Field Datu. Field data consisting of records of sow productivity were provided by the Hampshire Swine Registry, the American Landrace Association, and the National Spotted Swine Record. All sow productivity records included the following information: herd, contemporary group, sow, service sire, sire of sow, dam of sow, parity of sow, date of birth of litter, and NBA. Contemporary groups were defined by the breeders as a group of sows that were bred, gestated, and farrowed in a common herd, year, and seasonal time frame. Additional pedigree information used in the analysis included the sires and maternal grandsires of the service sire and sire of sow and the dams and paternal granddams of the dam of sow, if known. Hampshire and Landrace records were collected from 1985 to

2905 Downloaded from jas.fass.org by guest on July 13, 2011

2906

SEE ET AL.

1990, whereas Spotted records were from 1987 t o 1989. All records for NBA were adjusted for parity of sow using correction factors recommended by the National Swine Improvement Federation ( 1987). Data were edited to eliminate all records with missing information and records that were greater than four standard deviations from the mean. Also, records were removed when only one sire of sow was represented in a contemporary group. The number of records, contemporary groups, service sires, sires of sows, and dams of sows are presented in Table 1. The number of sows with NBA records were 7,103, 2,515, and 4,822 for Hampshire, Spotted, and Landrace, respectively. Model. Although an animal model that accounted for repeated records on sows, included all pedigree information, and accounted for random common environmental and service-sire effects would be the most appropriate, it becomes computationally prohibitive with large data sets. An alternative model was employed that reduced the number of equations to be solved by >70% for all breeds. The analysis model included all major effects and pedigree information for each animal effect, but not the sow itself, thereby allowing the analysis t o be performed with the available computing facilities. The analysis model was as follows: y = Xp + Zlul + Z2u2 + Z3ug + e, where y was the observation vector, 8 was the vector of fixed contemporary group effects, ul, ug, and ug were the vectors of random service sire, sire of sow, and dam of sow effects, and e was the vector of random residual effects. The X and Z matrices consisted on ones and zeros, relating the appropriate effects to the vectors of records. The mixed-model equations were written as follows:

variance of the service-sire effect, -0: = the variance of the sire of sow effect, 3; = the variance of the dam of sow effect, and 32 = the variance of the residual effects. The inverse numerator-relationship matrices for service sires Ai: and sires of sows Ai1 accounted for additive genetic relationships between themselves, their sires, and their paternal grandsires when known. The inverse numerator-relationship matrix Ai1 accounted for additive genetic relationships between dams of sows, their dams, and their maternal granddams. Expectations of the variances were as follows: E ( g J = CJ;, E ( e ) = %CT;, E ( 6 ) = %io$, + aAM+

4,

and E( expressed as a proportion of the total phenotypic variance were different across all the breeds reported. The Spotted breed ( . O 1) showed the smallest effect and the Hampshire breed ( . 0 5 ) the largest. These results suggest that maternal genetic influences may play different roles within each breed. These estimates do fit within the range reported in the literature and support the conclusion that the size of the litter in which a gilt was raised influences her later NBA records (Nelson and Robison, 1976; Rutledge, 1980). The proportion of the total variation for NBA due t o service sire in Hampshire (.02), Spotted (.Ol), and Landrace (.02) is in close agreement with previous studies. Both Mabry et al. ( 1988) and Feng ( 199 1) reported that the service sire accounted for 3% of the total variation in NBA, and Buytels and Long ( 199 1) found the service sire to be a significant effect, accounting for 1% of the total variation. Strang ( 1970) also reported a significant effect due to the service sire on NBA that accounted for .3% of the total variation. With a reported range of .3 to 3% of the total variation in NBA, it seems that the service sire has a very small but significant effect on NBA in swine. Van Vleck and Johnson (19801, in a study of the effect of service sire on dairy cattle milk production, found that the service sire accounted for 1%of the variation in milk yield and that the correlation between the service sire and sire of the cow was nearly zero. However, Van Vleck and Johnson (1980) concluded that the genetic and economic implications of the service-sire effect did not seem important. The service sire could influence NBA either by semen quality and quantity or by genetic effects that influence embryo development and survival. Further research is needed to investigate whether the servicesire effect is largely genetic or environmental, whether a covariance exists between the service sire and the sire of sow, whether direct selection on service sire can improve NBA, and whether improvements in accuracy of breeding value prediction can be made by including this effect in the prediction model.

Implications The heritability for number of pigs born alive in Hampshire, Spotted, and Landrace swine is low, approximately .13. Small but significant effects for maternal genetic and service-sire effects were found. Service sires were found to account for 1 to 2% of the total variation in number of pigs born alive. Further investigation is needed is determine whether the service-sire effect is primarily genetic or environmental. Production cost and productivity are linked to reproductive efficiency in commercial swine production. By capitalizing on additive genetic differences that are present, reproductive efficiency can be improved through selection.

Literature Cited Berger, P. J., G. R. Luecke, and J. A. Hoekstra. 1989. Iterative algorithms for solving mixed model equations. J. Dairy Sci. 72: 514. Buytels, J.A.A.M., and T. Long. 1991. The effect of service sire on pigs born alive. In: Proc. 9th Conf. Australian Assoc. of Anim. Breed. and Genet. p 303. June 24-27, 1991, Univ. of Melbourne, Victoria, Australia. Dempster, A. P., N. M. Laird, and D. B. Rubin. 1977. Maximum likelihood from incomplete data via the EM algorithm. J. R. Statist. Soc. Series B 39:l. Dong, M. C., L. D. Van Vleck, and G. R. Wiggans. 1988. Effect of relationships on estimation of variance components with a n animal model and restricted maximum likelihood. J. Dairy Sci. 71:3047. Falconer, D. S. 1989. Introduction to Quantitative Genetics (3rd Ed.). John Wiley & Sons, New York. Feng, X. 1991. Genetic parameter estimates for swine litter traits. Ph.D. Dissertation. Ohio State Univ., Columbus. Gu, Y., C. S. Haley, and R. Thompson. 1989. Estimates of genetic and phenotypic parameters of litter traits from closed lines of pigs. h i m . Prod. 49:477. Hageman, L. A., and D. M. Young. 1981. Applied Iterative Methods. Academic Press, New York. Henderson, C. R. 1986. Recent developments in variance and covariance estimation. J. h i m . Sci. 63:208. Kaplon, M. J., M. F. Rothschild, P. J . Berger, and M. Healey. 1991. Population parameter estimates for performance and reproductive traits in Polish Large White nucleus herds. J. Anim. Sci. 69:91. Kincaid, D. R., J . R. Respess, D. M. Young, and R. G. Grimes. 1982. Algorithm 586 ITPACK 2C: A FORTRAN package for solving large sparse linear systems by adaptive accelerated iterative methods. ACM Trans. Math. Software 8:302. Lamberson, W. G. 1990. Genetics parameters for reproductive traits.


VARIANCE COMPONENT ESTIMATES FOR LITTER SIZE In: Genetics of Swine. pp 75-83. USDA, Washington, DC. Mabry, J . W., B. K. Thomas, and M. N. McCarter. 1988. The effect of service sire on number born alive, number weaned and litter weaning weight in Yorkshire swine. J. Anim. Sci. 66(Suppl. 1): 16 (Abstr.). McCarter, M. N., Jr., J. W. Mabry, J. K. Bertrand, and L. L. Benyshek. 1987. Components of variance and covariance for reproductive traits in swine estimated from Yorkshire field data. J . Anim. Sci. 64:1285. Misztal, I. 1990. Restricted maximum likelihood estimation of variance components in animal model using sparse matrix inversion and a supercomputer. J. Dairy Sci. 73:163. National Swine Improvement Federation. 1987. Guidelines for Uniform Swine Improvement Programs (Rev. Ed.). USDA Program Aid 1157, USDA, Washington, DC. Nelson, R. E., and 0. W. Robison. 1976. Effects of postnatal maternal environment on reproduction of gilts. J. h i m . Sci. 43:71.

2909

Patterson, H. D., and R. Thompson. 1971. Recovery of interblock information when block sizes are unequal. Biometrics 58:545. Rutledge, J. J. 1980. Fraternity size and swine reproduction. I. Effect on fecundity of gilts. J . Anim. Sci. 51:868. SAS.1988. SASDML User’s Guide (Release 6.03). SAS Inst. Inc., Cary, NC. Searle, S. R. 1989. Variance components-some history and a summary of estimation methods. J. Anim. Breed. Genet. 106:l. Strang, G. S. 1970. Litter productivity in Large White pigs. Anim. Prod. 12:225. Strang, G. S., and J.W.B. King. 1970. Litter productivity in Large White pigs. Anim. Prod. 12:235. Strang, G. S., and C. Smith. 1979. A note on the heritability of litter traits in pigs. Anim. Prod. 28:403. Van Vleck, L. D., and L. P. Johnson. 1980. Genetic and economic implications of fetal effects on the dam. J. Dairy Sci. 63:1483.


Citations

This article has been cited by 8 HighWire-hosted articles: http://jas.fass.org/content/71/11/2905#otherarticles
