Research On Genomic Selection Technology For Beef Cattle Based On Parallel Computing

Genomic selection has been widely used for complex quantitative trait in farm animals.Estimations of breeding values for economical traits are most important to beef industry,it is worthwhile to investigate prediction accuracies of genomic selection for these traits in beef cattle.Firstly,we evaluated the accuracies of predictive genomic breeding values for longissimus dorsi muscle and tenderloin traits including live weight(LW),carcass weight(CW),longissimus dorsi muscle weight(LDW),tederloin weight(TW),chuck roll cap weight(CRW),sirloin weight(SW),rib eye roll cap weight(REW)and sum of LDW and TW(LTW)using Illumina BovineSNP770K Beadchip in 1217 Chinese Simmental beef cattle.Random masking strategy was used to form training sets and validation sets,and genomic breeding values were estimated using GBLUP,BayesA,BayesB and BayesC?.Correlation between adjusted phenotypic data and estimated breeding values and 5-fold cross validation were applied to evaluate the predictive results.We found the accuracies of genomic predictions ranged from 0.329(GBLUP for CRW)to 0.490(BayesB for SW).The average accuracies were 0.390(GBLUP),0.406(BayesA),0.432(BayesB)and 0.408(BayesC?),respectively.Bayesian methods were more accurate than GBLUP across most of traits.Genomic prediction based on Bayesian models via Markov Chain Monte Carlo(MCMC)is computationally intensive.Parallel computing is a type of computing architecture in which several processing units execute one computation task simultaneously to shorten running time.(1)we prosposed multiple chains parallel strategy for Bayesian methods based on traditionally parallel MCMC,we assessed genomic predictive abilities for traits of average daily gain weight(ADG),live weight(LW),carcass weight(CW),dressing percentage(DP),lean meat percentage(LMP)and retail meat weight(RMW)using simullation dataset and Simmental cattle real dataset.To evaluate and compare the abilities of prediction,marker effects were estimated using genomic BLUP(GBLUP)and three parallel Bayesian models,including multiple chains parallel BayesA,BayesB and BayesC?(PBayesA,PBayesB and PBayesC?).In simulation,same predictive accuracies between sequential Bayesian methods and parallel Bayesian methods,while running time of parallel Bayesian methods were reduced obviously.For real dataset,we found the accuracies of genomic predictions ranged from 0.195±0.084(GBLUP for LMP)to 0.424±0.147(PBayesB for CW)in Chinese Simmental cattle.The average accuracies across all traits were 0.327±0.085(GBLUP),0.335±0.063(PBayesA),0.347±0.093(PBayesB)and 0.334±0.077(PBayesC?).(2)To obtain an accurate result of genomic prediction,the burn-in of MCMC implemented in Bayesian model is always set to a large value,however,more burn-in iteration can't improve the performance of MCMC when it reaches equilibrium.We proposed an automatically tuned strategy for setting of burn-in value in multiple chains parallel MCMC for genomic prediction,in which multiple chains convergence diagnosis was used to determine burn-in value of MCMC.We utilized BayesA and BayesC? in the studies of genomic predictions,and compared tuned burn-in multiple chains parallel BayesA(TunBpBayesA),tuned burn-in multiple chains parallel BayesC?(TunBpBayesC?)with fixed burn-in multiple chains parallel BayesA(FixBpBayesA),fixed burn-in multiple chains parallel BayesC?(FixBpBayesC?)and GBLUP in simulation.Predictive accuracies of TunBpBayesA(or TunBpBayesC?)were same to those of FixBpBayesA(or FixBpBayesC?),while speedup ratios of TunBpBayesA(or TunBpBayesC?)were higher than those of FixBpBayesA(or FixBpBayesC?).Moreover,using 1217 real data of Chinese Simmental beef cattle genotyped with Illumina Bovine 770K SNP BeadChip,we found TunBpBayesC? performed better than TunBpBayesA and GBLUP for four traits using five-folds cross validations.Our results indicated:?The estimation of direct genomic breeding values can offer important insights into improving the accuracy of predicting genetic merit of animals at young ages and may help to identify superior candidates for tender beef loin in breeding program.?genomic selection using multiple chains parallel Bayesian models are feasible for slaughter traits in Chinese Simmental cattle.? Genomic prediction using tuned burn-in strategies can achieve higher speedup than those of fixed burn-in strategies in multiple chains parallel BayesA and BayesC?,and it exhibits good scalability for usage of CPU cores and flexibility for extensions of Bayesian models.