Acurácia de predição do valor genético genômico em bovinos da raça Gir para produção de leite até os 305 dias de lactação
Carregando...
Arquivos
Data
2023-10-27
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
— Estudos para a incorporação de dados moleculares em conjunto com a avaliação genética animal têm sido propostos com os objetivos de aumento da acurácia na predição de valores genéticos genômicos (GEBV), aumento da intensidade de seleção e diminuição dos intervalos de gerações, visando aumentar o ganho genético pela seleção ao longo de sucessivas gerações de acasalamento. O uso de painéis customizados de nucleotídeos de polimorfismo único (SNP) de baixa densidade se mostra interessante devido ao menor custo em relação aos painéis de SNP de alta densidade. Dessa forma, um maior número de animais candidatos a seleção podem ser genotipados. Para a seleção dos SNP que compõem os painéis customizados o algoritmo de aprendizado de máquina tem demonstrado bons resultados em pesquisa.
No entanto, com painéis de baixa densidade de SNP, dificuldades de convergência podem ocorrer nas análises de predição do GEBV. Desse modo, o objetivo desse estudo foi verificar a habilidade de predição do GEBV em bovinos da raça Gir a partir de subconjunto de SNP, classificado segundo sua importância para a produção de leite até os 305 dias de lactação. Os SNP foram classificados por regressão não paramétrica por meio do algoritmo Random Forest (RF) e foram testados 20 cenários, os quais variaram quanto à proporção da matriz A22 dentro da G e A22 -1 dentro da matriz H -1 , buscando-se verificar o impacto do subconjunto de SNP na acurácia e no viés de predição. A presença de genes em uma janela de 1Mb foi verificada usando os pacotes biomaRt, com a versão 110 de Ensembl Genomes, clusterProfiler do software R. Dentre os cenários de RF, o melhor foi aquele em que os grupos de treinamento e validação foram separados de acordo com as gerações, cujo Ntree e Mtry foram 8192 e 0,3 vezes no número de SNP presentes no painel de 400 mil marcadores, respectivamente. Os SNP com maior relevância apresentaram oito genes relacionados a importantes vias metabólicas e funções biológicas, presentes em várias espécies de animais. Para os diferentes cenários testados observou-se diferenças significativas (p-valor < 0,05) entre estimativas da variância e da média do GEBV, do fenótipo predito (YHat) e do valor dos resíduos que representam o valor fenotípico descontado dos efeitos incluídos no modelo) para os diferentes cenários de ponderação da matriz G e A22 -1 testados e para os diferentes painéis de customizados. Esse resultado mostrou-se promissor, uma vez que é possível criar painéis de SNP com menores densidades sem perder o poder de predição do GEBV.
Studies for the incorporation of molecular data in conjunction with animal genetic evaluation have been proposed with the objectives of increasing the accuracy in predicting genomic genetic values (GEBV), increasing the intensity of selection and decreasing generation intervals, aiming to increase the gain genetic by selection over successive generations of mating. The use of customized low-density single nucleotide polymorphism (SNP) panels is interesting due to the lower cost compared to high-density SNP panels. In this way, a greater number of candidate animals for selection can be genotyped. For the selection of SNPs that make up customized panels, the machine learning algorithm has demonstrated good research results. However, with low SNP density panels, convergence difficulties may occur in GEBV prediction analyses. Therefore, the objective of this study was to verify the ability to predict GEBV in Gir cattle based on a subset of SNP, classified according to their importance for milk production up to 305 days of lactation. The SNPs were classified by non-parametric regression using the Random Forest (RF) algorithm and 20 scenarios were tested, which varied in terms of the proportion of the matrix "A" _"22" within G and "A" _"22" ^ "-1" within the matrix "H" ^"-1", seeking to verify the impact of the SNP subset on accuracy and prediction bias. The presence of genes in a 1Mb window was verified using the biomaRt packages, with version 110 of Ensembl Genomes, clusterProfiler of the R software. Among the RF scenarios, the best was the one in which the training and validation groups were separated from according to generations, whose Ntree and Mtry were 8192 and 0.3 times the number of SNP present in the panel of 400 thousand markers, respectively. The most relevant SNPs presented eight genes related to important metabolic pathways and biological functions, present in several animal species. For the different scenarios tested, significant differences (p-value < 0.05) were observed between estimates of the variance and mean of GEBV, the predicted phenotype (YHat) and the value of the residues that represent the discounted phenotypic value of the effects included in the model) for the different matrix weighting scenarios G and "A" _"\2\2"^"\-\1" tested and for the different customized panels. This result was promising, since it is possible to create SNP panels with lower densities without losing the predictive power of GEBV.
Studies for the incorporation of molecular data in conjunction with animal genetic evaluation have been proposed with the objectives of increasing the accuracy in predicting genomic genetic values (GEBV), increasing the intensity of selection and decreasing generation intervals, aiming to increase the gain genetic by selection over successive generations of mating. The use of customized low-density single nucleotide polymorphism (SNP) panels is interesting due to the lower cost compared to high-density SNP panels. In this way, a greater number of candidate animals for selection can be genotyped. For the selection of SNPs that make up customized panels, the machine learning algorithm has demonstrated good research results. However, with low SNP density panels, convergence difficulties may occur in GEBV prediction analyses. Therefore, the objective of this study was to verify the ability to predict GEBV in Gir cattle based on a subset of SNP, classified according to their importance for milk production up to 305 days of lactation. The SNPs were classified by non-parametric regression using the Random Forest (RF) algorithm and 20 scenarios were tested, which varied in terms of the proportion of the matrix "A" _"22" within G and "A" _"22" ^ "-1" within the matrix "H" ^"-1", seeking to verify the impact of the SNP subset on accuracy and prediction bias. The presence of genes in a 1Mb window was verified using the biomaRt packages, with version 110 of Ensembl Genomes, clusterProfiler of the R software. Among the RF scenarios, the best was the one in which the training and validation groups were separated from according to generations, whose Ntree and Mtry were 8192 and 0.3 times the number of SNP present in the panel of 400 thousand markers, respectively. The most relevant SNPs presented eight genes related to important metabolic pathways and biological functions, present in several animal species. For the different scenarios tested, significant differences (p-value < 0.05) were observed between estimates of the variance and mean of GEBV, the predicted phenotype (YHat) and the value of the residues that represent the discounted phenotypic value of the effects included in the model) for the different matrix weighting scenarios G and "A" _"\2\2"^"\-\1" tested and for the different customized panels. This result was promising, since it is possible to create SNP panels with lower densities without losing the predictive power of GEBV.
Descrição
Palavras-chave
Genética, Lactação, Bovinos, Melhoramento genético
Como citar
SENA, T. M. - Acurácia de predição do valor genético genômico em bovinos da raça Gir para produção de leite até os 305 dias de lactação - 2023, 69f - Tese (Doutorado em Genética e Melhoramento Animal) - Universidade Estadual Paulista, Jaboticabal, 2024.