Prediction of Genetic Values of Sugarcane Using Genomic Information

Successful applications of genomic prediction in animal and plant science have encouraged its use in plant breeding. The development of genomic tools for accelerate sugarcane breeding has been delayed compared to other major crops; therefore, empirical studies must be conducted to evaluate the usefulness of this approach for selection of this complex genome crop. A training population of 514 clones conducted in a second phase trial was genotyped with 37,024 SNP markers. Genetic values were predicted for four traits (percentage of sucrose in sugarcane – PC, tonnes of stalks per hectare – TSH, tonnes of pol per hectare – TPH and percentage of fiber in bagasse – FB) using five statistical predictive models: Genomic BLUP – GBLUP, Bayesian LASSO – BL, Bayes A – BA, Bayes B – BB and Bayesian Ridge Regression – BRR. The accuracies of these methods were assessed by a 10-fold cross-validation procedure. There were no major differences in accuracies among methods, but a marked difference among traits. Accuracies values varied from 0.412 to 0.419 for PC, 0.220 to 0.239 for TSH, 0.068 to 0.081 for TPH and 0.561 to 0.570 for FB, suggesting that yield components (TSH and TPH) seem to need higher marker coverage. Although phenotypic selection exhibited higher accuracies than genomic selection, the results achieved show that molecular markers could be used for prediction of genetic merit of clones, shortening the time of current selection, which requires large field experiments. Thus, the use of genomic information for improvement of desirable agronomic traits in sugarcane seems to be promising. (FAPEMIG, CNPq, RIDESA,CAPES)