Genome wide selection optimization in maize breeding

• Ano de defesa: 2023
• Autor(a) principal: Bernardeli, Arthur Martins Almeida
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa; Genética e Melhoramento
• Palavras-chave em Português: Milho - Melhoramento genético; Milho - Seleção; Marcadores genéticos; Milho híbrido; Melhoramento Vegetal
• Link de acesso: https://locus.ufv.br//handle/123456789/30839; https://doi.org/10.47328/ufvbbt.2023.173

Maize is a staple crop and the most grown cereal worldwide. The expansion of this crop was possible due to efforts in management and breeding. In the breeding standpoint, advances were achieved in the release of hybrids presenting heterosis, field experimental design and analyses, establishment of heterotic patterns, and effective seed production and marketing. From the last decade on, advances in data analyses benefited from the surge of genotypic data, allowing the prediction of hybrids without being tested through genomic selection approaches. This study aims to convert a high-density SNP data set and use it in a genomic selection or predicting non- tested hybrids and non-observed environments, and for indicating most promising mating parent material for obtaining hybrids and inbred lines for ASI, EPP, FFT, GY, and MFT maize traits. For that, we ranked the SNPs according to their effects from a ME analyses and selected the minimum portion of markers that reached the plateau of prediction accuracy per chromosome, followed by eliminating the repeated markers between traits, and removing the ones tightly linked according to LD analyses. For the GS of hybrids and environments, three methods that comprised GCA and SCA main and interaction effects were fitted, and the prediction accuracy was assessed. The step of selecting parent material was performed according to PS, GS, and GM. The GM methods used the marker effects predicted in the previous GS step, and the 40 top- and bottom-performing crosses and their respective parent lines were selected for each trait. The selected SNPs maintained the accuracy for all traits under drought or well-watered conditions when compared to using full SNP set. For GWS of hybrids, Model 3 performed better for all traits when cross validation schemes had information of all environments (CV1 and CV2) in terms of prediction accuracy, and Model 2 performed better when there was missing information about environments (CV0 and CV00). The mating parents chosen for positive selection were different than the ones from negative selection, ensuring maximization of gains for hybrid and inbred lines development. The highest coincidences of selected parent lines occurred in GS-based methods (Methods 1, 3, 5, 7, 9, 11, 13, and 15), where parents were directly selected based on means or GCA/SCA (and interaction) values of their respective hybrids. The methods based on crosses simulations (Methods 2, 4, 6, 8, 10, 11, 12, 14, and 16) had moderate to low coincidences, but were consistent in indicating the best parent materials overall. GS- and GM-based parent selection results must be further compared to Method 17 (observed crosses) for an effective validation. PS, GS, and GM methods together must help in the decision making of selecting parent material for future crosses. These approaches must be further performed using other training populations. Keywords: Cross-validation. Hybrids. Inbred Lines. Prediction Accuracy. SNP Markers.