%0 Journal Article %1 Hay201572 %A Hay, El Hamidi %A Rekaya, Romdhane %D 2015 %J Livestock Science %K 2015 genomics methodology %P 72 - 76 %R http://dx.doi.org/10.1016/j.livsci.2015.10.009 %T A structural model for genetic similarity in genomic selection of admixed populations %U http://www.sciencedirect.com/science/article/pii/S187114131530010X %V 181 %X Abstract Current approaches for dealing with admixed and crossbred populations in genomic selection rely on using different groups of animals in training sets. These approaches benefit from increased power as a result of increasing the size of the training set. However, the performance largely depends on the genetic similarity between the sub-populations of the admixed population. Our proposed multi-compartment model where the effect of an \SNP\ could be different between breeds and parameterized as a function of its effect on one of the breeds in admixed population through a one to one mapping function, was able to remediate some problems of the pooled data approaches but still suffers from the high dimensionality of the unknown parameters to estimate. To overcome this problem, we propose not to estimate the mapping parameter α for each \SNP\ but rather to build a model for α as a function of information already available in the genotype data via a hierarchical structural model. In this study, α was modeled as a function of the change in linkage disequilibrium. An admixed population (A and B) and crossbred populations (AB, BA, BxAB, BxBAB) were simulated. Individuals were genotyped for 300 \SNPs\ and measured for a quantitative trait with 0.5 heritability. Three analyses were conducted: 1) classical pooled data (M1); 2) pooled data using the multi-compartment model and α for each \SNP\ (M2); and 3) pooled data using multi-compartment model and our structural model for α (M3). The accuracy of (M1) tended to be much lower than using models \M2\ or M3. The prediction accuracies for line A using model \M1\ was 0.39 compared to 0.56 and 0.43 using \M2\ and M3, respectively. The accuracies using the structural model (M3) resulted in intermediate to those obtained using \M1\ and M2. The relatively good performance obtained using \M2\ indicates that it is possible to model α as a function of the information already available in the genotype data and to substantially reduce the number of parameters to be estimated. Additionally, the model for α could be further refined which will likely lead to better performances.

@article{Hay201572, abstract = {Abstract Current approaches for dealing with admixed and crossbred populations in genomic selection rely on using different groups of animals in training sets. These approaches benefit from increased power as a result of increasing the size of the training set. However, the performance largely depends on the genetic similarity between the sub-populations of the admixed population. Our proposed multi-compartment model where the effect of an \{SNP\} could be different between breeds and parameterized as a function of its effect on one of the breeds in admixed population through a one to one mapping function, was able to remediate some problems of the pooled data approaches but still suffers from the high dimensionality of the unknown parameters to estimate. To overcome this problem, we propose not to estimate the mapping parameter α for each \{SNP\} but rather to build a model for α as a function of information already available in the genotype data via a hierarchical structural model. In this study, α was modeled as a function of the change in linkage disequilibrium. An admixed population (A and B) and crossbred populations (AB, BA, BxAB, BxBAB) were simulated. Individuals were genotyped for 300 \{SNPs\} and measured for a quantitative trait with 0.5 heritability. Three analyses were conducted: 1) classical pooled data (M1); 2) pooled data using the multi-compartment model and α for each \{SNP\} (M2); and 3) pooled data using multi-compartment model and our structural model for α (M3). The accuracy of (M1) tended to be much lower than using models \{M2\} or M3. The prediction accuracies for line A using model \{M1\} was 0.39 compared to 0.56 and 0.43 using \{M2\} and M3, respectively. The accuracies using the structural model (M3) resulted in intermediate to those obtained using \{M1\} and M2. The relatively good performance obtained using \{M2\} indicates that it is possible to model α as a function of the information already available in the genotype data and to substantially reduce the number of parameters to be estimated. Additionally, the model for α could be further refined which will likely lead to better performances. }, added-at = {2016-05-15T19:19:51.000+0200}, author = {Hay, El Hamidi and Rekaya, Romdhane}, biburl = {https://www.bibsonomy.org/bibtex/2b93f2dc37062fbcdb26cb1dc27305a2a/uga.abgg}, doi = {http://dx.doi.org/10.1016/j.livsci.2015.10.009}, interhash = {ad73de034243be25f230646e52f52963}, intrahash = {b93f2dc37062fbcdb26cb1dc27305a2a}, issn = {1871-1413}, journal = {Livestock Science }, keywords = {2015 genomics methodology}, pages = {72 - 76}, timestamp = {2016-05-15T19:19:51.000+0200}, title = {A structural model for genetic similarity in genomic selection of admixed populations }, url = {http://www.sciencedirect.com/science/article/pii/S187114131530010X}, volume = 181, year = 2015 }