The Study And Application Of Parameters Optimization Of Bayesian Methods For Genomic Selection In Livestock

Genomic selection refers to use the whole genome markers to estimate the breedingvalues, which was called genomic estimated breeding values (GEBV). At present, genomicselection has become an important technology in livestock breeding. Various methods havebeen proposed and more efficient models are always being developed. However, it is ofgreat significance to exploit better models and algorithm for improving the accuracy ofgenomic selection. In this paper, a detailed overview of Bayesian methods in livestockgenomic selection will be introduced. Two strategies were used to optimize the parameterof degree of freedom and scale parameter in Bayesian methods for genomic selection.First, the degree of freedom and scale parameter of scaled inverse chi-squaredistribution in BayesB and BayesC π methods were optimized. The research wasoptimized and improved on the basis of BayesB and BayesCπ methods. According to thehierarchy Bayesian theory, the degree of freedom and scale parameter in scaled inversechi-square distribution of marker effect variance were optimized, which were not presetfixed value and needed to be calculated in each round of iteration. Finally the parametersaccording with the experimental group were drawn out, in order to improve the accuracy ofthe genomic estimated breeding value. In view of the improving methods of BayesB andBayesCπ, they were named as BayesFB and BayesFCπ. All the programs used in thisstudy were compiled by C language. Two parts of simulation data set were used to verifythe reliability of BayesFB and BayesFCπ. First part: GPOPSIM program was used tosimulate phenotype and genotype. Different scenarios were simulated to study withdifferent QTL number (50,200,500,1000) and different heritability (0.05,0.2,0.5,0.9),and each simulation was repeat10times to verify the simulation program reliability.Second part: QTL-MAS2011was used to verify the reliability of new program. BayesA,BayesB and BayesCπ were used to compare the accuracy of genomic estimated breedingvalue with BayesFB and BayesFCπ. The results show that for two simulation data sets,the accuracy of genomic estimated breeding value of BayesFB and BayesFCπ wasrespectively higher than BayesB and BayesCπ.Second, BayesA method is optimized to regain the prior of degree of freedom andscale parameter, discussed below. Different traits were assumed that they had specificdegree of freedom and the scale parameter. RRBLUP and GAPIT programs were firstly used to regain the two parameters and then putting these two parameters into BayesAmethod, which was called BayesPA. Simulation data reveal that different degree offreedom and scale parameter will influence the accuracy of genomic estimated breedingvalue. Compared with BayesA, BayesPA had higher accuracy.Third,824Chinese beef Simmental cattle were used to calculate the genomicestimated breeding value by BayesA, BayesB, BayesC π,BayesFB, BayesFC π andBayesPA methods and compare the accuracy of GEBVs for five important economiccharacters for evaluating the performance of beef cattle production traits including liveweight, carcass, tenderloin, dressing percentage and pure meat percentage. The markereffect maps of live weight, carcass and tenderloin weight were made. The results displayedthat the accuracy of GEBVs estimated by BayesFB and BayesFCπ methods were notimproved. However, BayesPA comparing to BayesA method improved significantly.