Effect of Allele Dosage in the Accuracy of Genomic Selection in Guinea Grass, an Autotetraploid Forage

Genomic selection (GS) is an effective method to perform predictions of genomic breeding values of genotypes evaluated in breeding programs. However, most of the available GS models were developed for diploids and still is not well established for polyploids species. Here, we investigate the importance of considering alllele dosage in GS models in autopolyploid species, with applications in Panicum maximum. The population was composed by 530 half-sibs from 19 families. Adjusted means were obtained via mixed models analyses from an augmented block design experiment evaluating organic matter of leaf (OML) and cellulose of stem (CS). Genotypic data from genotyping-by-sequencing (GBS) read depths were obtained in two ways: diploid dosage (DD) using TASSEL-GBS pipeline, and tetraploid dosage (TD) using SuperMASSA software. Markers with less than 5% missing data were selected and imputed using random sampling and considering the dosage class proportion for each marker. Six genomic models were analyzed using R packages sommer (G-BLUP) and BGLR (BRR, BLASSO, Bayes A, Bayes B, and Bayes C). Cross validation was conducted considering family information. For both traits, the use of allele dosage increased the accuracy of GS models. The highest accuracy for OML was 0.7376, with Bayes A and TD, and the lowest was 0.6717, with G-BLUP and DD. Considering CS, the highest accuracy was 0.5470, with BRR and TD, and the lowest was 0.3591, with G-BLUP and DD. In conclusion, polyploid allele dosage increases the accuracy of GS in guinea grass and shows its potential use in breeding programs of autotetraploid species.