Bayesian methods applied to genomic prediction of milk production and quality traits of milk buffaloes

• Grant number: 15/18614-0
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Effective date (Start): January 15, 2016
• Effective date (End): September 14, 2016
• Field of knowledge: Agronomical Sciences - Animal Husbandry - Genetics and Improvement of Domestic Animals
• Principal Investigator: Humberto Tonhati
• Grantee: Camila da Costa Barros
• Supervisor: Guilherme Jordão de Magalhães Rosa
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil
• Research place: University of Wisconsin-Madison (UW-Madison), United States
• Associated to the scholarship: 13/24427-3 - Genomic selection for yield and quality of milk buffalo, BP.DR

Abstract

Buffalo milk is characterized by high amount in terms of fat, protein and total solids compared to milk from other animal species, allowing higher yields of its derivatives, especially in the manufacture of mozzarella cheese. Genomic selection provides benefits to the animal breeding due to the direct use of DNA information in the selection process, increasing the genetic gain compared to traditional selection based only on the phenotype and pedigree. The objective of this study will evaluate and compare different models in predicting capacity of genomic breeding values obtained through different methodologies of Bayesian alphabet with a view to genomic selection for milk production traits and the percentages of fat and protein from buffalo milk. Phenotype and genotype information from 452 animals will be used (57 males and 395 females) of Murrah buffaloes, from twelve herds in the State of São Paulo. The animals are already genotyped with specific chip for buffaloes, which identifies 90,000 SNPs spread on the genome buffalo (Axiom® Buffalo Genotyping Array). Deregressed breeding values (dEBV) will be calculated and used as response variables in genomic analysis. Genomic values will be predicted using the following methods: Bayes A, Bayes B, Bayes C? and Improved Bayesian LASSO. The Bayesian models will be compared using cross-validation performed separately for each trait analyzed. The sample will be divided into two groups: the training set, which will be used for estimation of SNP marker effects, and validation set, which will be used to evaluate the correlation between predicted breeding values through estimates from the training population and corrected phenotypes observed. From the Pearson correlation between the predicted values in the validation and observed in the data set it will be possible to obtain the accuracy of prediction. (AU)