Predictive methods using pedigrees, markers, and images for the genetic improvement of sugarcane

• Ano de defesa: 2024
• Autor(a) principal: Gonçalves, Mateus Teles Vital
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa; Estatística Aplicada e Biometria
• Palavras-chave em Português: Cana-de-açúcar - Melhoramento genético; BLUP; Poliploidia; Saccharum; Genômica; Cana-de-açúcar - Seleção; Modelos lineares (Estatística); Estatística Aplicada e Biometria
• Link de acesso: https://locus.ufv.br/handle/123456789/33830; https://doi.org/10.47328/ufvbbt.2024.652

The overall objective of this thesis was to assess predictive methods that leverage different sources of information, to improve genetic evaluation of sugarcane genotypes. The plant material and experimental data used consisted of early-generation field trials conducted by the genetic breeding program of the Universidade Federal de Viçosa (PMGCA). In the first chapter I discussed the content, introducing research questions and the main objectives of this thesis. In the second chapter, we compared pedigree- based best linear unbiased prediction (PBLUP), genomic-based (GBLUP), and single- step (ssGBLUP) models for the genetic evaluation of a sugarcane population in which only a subset of individuals was genotyped. Models were evaluated in two cross- validation (CV) schemes: validation using genotyped individuals (CV1) and validation using nongenotyped individuals (CV2). Our results suggest that genetic evaluation us- ing the ssGBLUP models may be an alternative approach for sugarcane. Also, results showed that models including only pedigree information gave relatively high prediction accuracies, suggesting that pedigrees are an important source of genetic information, particularly for sugarcane and other crop species with complex polyploid genomes. In the third chapter, we evaluated the integration of unoccupied aerial systems (UAS)- based red-green-blue (RGB) imaging with pedigree and genomic prediction models to improve selection accuracy for tonnes of cane per hectare (TCH). The objectives were to estimate genetic parameters and trends for TCH and RGB-image extracted traits, and to compare the performance of single-trait with multi-trait genomic and pedigree prediction models that incorporate RGB-image extracted traits. The performance of models was evaluated in terms of Pearson’s correlation between adjusted and predicted phenotypes, and mean squared error (MSE) using three cross-validation schemes, which varied in the level of phenotypic information available: ST, without secondary traits; MT-1, secondary traits in the training set; and MT-2, secondary traits in both, training and testing sets. We used data of an augmented block design trial, consisting of 385 clones. Clones were phenotyped at the second ratoon stage for TCH, and for 12 RGB-image extracted traits collected in a single flight. In general, we found low genetic correlation between TCH and RGB-image extracted traits, and moderate narrow-sense heritability estimates for RGB-image extracted traits. Overall indirect response to selection of RGB-image extracted traits was higher compared to direct response to selection for TCH. Our results suggest that accuracies of multi-trait models that incorporated RGB- image extracted traits did not improve compared to single-trait models for predicting TCH. Future research should investigate alternative sensor technologies and optimize UAS-based data collection. Keywords: RGB; Single-step models; Multi-trait models; Polyploid; Saccharum spp.; Genomic prediction.