Impacto de estratégias de ponderação de marcadores na seleção genômica
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidade Federal de Viçosa
Abstract
O avanço da seleção genômica ampla (Genome-Wide Selection – GWS) tem pro- porcionado ganhos expressivos em programas de melhoramento genético ao permi- tir a predição de valores genéticos com base em marcadores moleculares distribuí- dos ao longo do genoma. O método Genomic Best Linear Unbiased Prediction (G- BLUP) é um dos mais utilizados em GWS devido à sua eficiência computacional e à capacidade de lidar com a alta dimensionalidade dos dados. Entretanto, o G-BLUP assume contribuição homogênea dos marcadores na construção da matriz de pa- rentesco genômico, desconsiderando diferenças na importância dos marcadores e sua associação com os loci causais. Nesse contexto, este estudo teve como objetivo avaliar o impacto de diferentes estratégias de ponderação de marcadores molecula- res sobre parâmetros genéticos e estatísticos ao longo de ciclos de seleção genômi- ca recorrente. Foram considerados pesos derivados de estudos de associação ge- nômica ampla (GWAS), do método BayesB e de medidas de associação entre mar- cadores, como desequilíbrio de ligação (LD), LD-score e fator de inflação da variân- cia (VIF). Dados simulados foram gerados considerando diferentes arquiteturas ge- néticas e níveis de herdabilidade em um programa com 30 ciclos de seleção recor- rente. O desempenho das estratégias foi avaliado por meio da acurácia preditiva, viés, concordância dos melhores indivíduos, ganho genético, diversidade genética e endogamia. Os resultados demonstraram que a eficiência das estratégias de ponde- ração depende da arquitetura genética da característica e da métrica avaliada, não havendo uma abordagem universalmente superior ao G-BLUP tradicional. De modo geral, a ponderação exerceu pouca influência sobre o ganho genético, mas favore- ceu, em determinados cenários, maior preservação da diversidade genética e menor endogamia. Estratégias que combinaram medidas de importância dos marcadores com informações de associação entre marcadores apresentaram os resultados mais consistentes. Conclui-se que a ponderação da matriz de parentesco genômico cons- titui uma estratégia complementar ao G-BLUP tradicional, cuja adoção deve consi- derar os objetivos específicos do programa de melhoramento, especialmente quan- do se busca conciliar resposta à seleção e manutenção da variabilidade genética. A dissertação está organizada em duas seções. A primeira apresenta a fundamentação teórica da seleção genômica, com ênfase no modelo G-BLUP, nas estratégias de ponderação de marcadores, nas medidas de associação entre marcadores e nas métricas utilizadas para avaliação dos modelos. A segunda descreve, por meio de dados simulados, a avaliação comparativa de diferentes estratégias de ponderação da matriz de parentesco genômico sobre parâmetros genéticos e estatísticos ao longo de ciclos de seleção genômica recorrente. Palavras-chave: G-BLUP; seleção genômica recorrente; arquitetura genética; ponderação de marcadores; matriz de parentesco genômica.
The advancement of Genome-Wide Selection (GWS) has led to substantial genetic gains in breeding programs by enabling the prediction of genomic estimated breeding values using molecular markers distributed across the genome. Among the available methods, Genomic Best Linear Unbiased Prediction (G-BLUP) is one of the most widely adopted due to its computational efficiency and its ability to handle high- dimensional genomic data. However, G-BLUP assumes that all markers contribute equally to the construction of the genomic relationship matrix, disregarding differ- ences in marker importance and their association with causal loci. In this context, the present study aimed to evaluate the impact of different molecular marker weighting strategies on genetic and statistical parameters throughout recurrent genomic selec- tion cycles. Marker weights were derived from Genome-Wide Association Studies (GWAS), the BayesB method, and marker association measures, including linkage disequilibrium (LD), LD-score, and the variance inflation factor (VIF). Simulated da- tasets were generated under different genetic architectures and heritability levels in a breeding program comprising 30 recurrent selection cycles. The performance of the weighting strategies was assessed based on predictive accuracy, bias, agreement among the top-ranked individuals, genetic gain, genetic diversity, and inbreeding. The results showed that the effectiveness of marker weighting strategies depends on both the genetic architecture of the trait and the evaluation metric, with no single ap- proach consistently outperforming the traditional G-BLUP. Overall, marker weighting had little effect on genetic gain but promoted greater genetic diversity and lower in- breeding under specific scenarios. Strategies combining marker importance measures with marker association information yielded the most consistent results. These findings indicate that weighting the genomic relationship matrix should be re- garded as a complementary strategy to the traditional G-BLUP, and its adoption should be guided by the specific objectives of the breeding program, particularly when the goal is to balance selection response with the maintenance of genetic vari- ability. This dissertation is organized into two sections. The first presents the theoret- ical background of genomic selection, with emphasis on the G-BLUP model, marker weighting strategies, marker association measures, and the metrics used to evaluate genomic prediction models. The second describes, through simulated data,a comparative evaluation of different genomic relationship matrix weighting strategies on genetic and statistical parameters throughout recurrent genomic selection cycles. Keywords: G-BLUP; recurrent genomic selection; genetic architecture; marker weighting strate-gies; genomic relationship matrix.
The advancement of Genome-Wide Selection (GWS) has led to substantial genetic gains in breeding programs by enabling the prediction of genomic estimated breeding values using molecular markers distributed across the genome. Among the available methods, Genomic Best Linear Unbiased Prediction (G-BLUP) is one of the most widely adopted due to its computational efficiency and its ability to handle high- dimensional genomic data. However, G-BLUP assumes that all markers contribute equally to the construction of the genomic relationship matrix, disregarding differ- ences in marker importance and their association with causal loci. In this context, the present study aimed to evaluate the impact of different molecular marker weighting strategies on genetic and statistical parameters throughout recurrent genomic selec- tion cycles. Marker weights were derived from Genome-Wide Association Studies (GWAS), the BayesB method, and marker association measures, including linkage disequilibrium (LD), LD-score, and the variance inflation factor (VIF). Simulated da- tasets were generated under different genetic architectures and heritability levels in a breeding program comprising 30 recurrent selection cycles. The performance of the weighting strategies was assessed based on predictive accuracy, bias, agreement among the top-ranked individuals, genetic gain, genetic diversity, and inbreeding. The results showed that the effectiveness of marker weighting strategies depends on both the genetic architecture of the trait and the evaluation metric, with no single ap- proach consistently outperforming the traditional G-BLUP. Overall, marker weighting had little effect on genetic gain but promoted greater genetic diversity and lower in- breeding under specific scenarios. Strategies combining marker importance measures with marker association information yielded the most consistent results. These findings indicate that weighting the genomic relationship matrix should be re- garded as a complementary strategy to the traditional G-BLUP, and its adoption should be guided by the specific objectives of the breeding program, particularly when the goal is to balance selection response with the maintenance of genetic vari- ability. This dissertation is organized into two sections. The first presents the theoret- ical background of genomic selection, with emphasis on the G-BLUP model, marker weighting strategies, marker association measures, and the metrics used to evaluate genomic prediction models. The second describes, through simulated data,a comparative evaluation of different genomic relationship matrix weighting strategies on genetic and statistical parameters throughout recurrent genomic selection cycles. Keywords: G-BLUP; recurrent genomic selection; genetic architecture; marker weighting strate-gies; genomic relationship matrix.
Description
Citation
PEREIRA, Guilherme Flaviano. Impacto de estratégias de ponderação de marcadores na seleção genômica. 2026. 71 f. Dissertação (Mestrado em Estatística Aplicada e Biometria) - Universidade Federal de Viçosa, Viçosa. 2026.
