Integrating NIR and genomic data for predicting fiber and sucrose content in sugarcane

• Ano de defesa: 2019
• Autor(a) principal: Gonçalves, Mateus Teles Vital
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Espectroscopia de infravermelho; Plantas - Melhoramento genético; Melhoramento Vegetal
• Link de acesso: https://locus.ufv.br//handle/123456789/28702

The main goal of this dissertation was to investigate candidate methodologies to circumvent some of the bottlenecks of the sugarcane genetic breeding program of the Universidade Federal de Viçosa (PMGCA). In chapter one, we developed regression and classification models using near-infrared spectroscopy to predict and classify sugarcane clones based on two feedstock quality parameters. The values measured by reference methods and predicted by PLS and PLS-DA models were compared. The PLS models developed had moderate accuracies. The correlation coefficients of prediction obtained were: 0.732 for fibre content and 0.665 for sucrose content. The PLS-DA models built to classify clones based on PC% showed the ideal value of 1 for sensitivity, whereas models based on FIB% showed a moderate value of 0.758. Both models exhibited similar classification errors: 0.185 and 0.195 for FIB% and PC%, respectively. These results indicate the feasibility of NIR spectroscopy coupled with multivariate analysis for the substitution of current time-consuming methods in the evaluation of large populations of sugarcane clones. In chapter two, we investigate whether the accuracy of genomic selection is improved, by combining the NIR spectra matrix to a SNP genotyping matrix into a single regression analysis. The accuracy of genomic selection models was evaluated using the Kennard-Stone algorithm and computing the correlation between the breeding values obtained using phenotypic measurements and breeding values estimated using genomic information. Combining the NIR spectroscopy information to the genomic dataset improved the correlation coefficient estimates for FIB% and PC%. The results found in this study suggest that models including NIR spectra-derived data coupled with molecular markers information resulted in higher predictive ability. Hence, this approach could be used to enhance the efficiency of selection of sugarcane clones by reducing breeding time cycles and thus, increase genetic gains at the PMGCA.