Genomic prediction of ordinal traits via application of the BLASSO and BayesCπ methods in simulated data

Geraldo Magela Da Cruz Pereira,Andrew De Paula Ribeiro,Sebastião Martins Filho

doi:10.33837/msj.v3i1.1112

Geraldo Magela Da Cruz Pereira, Andrew De Paula Ribeiro + Show 1 more

Open Access

https://doi.org/10.33837/msj.v3i1.1112

Copy DOI

Abstract

This paper aims at evaluating the use of BLASSO and BayesCπ methods for the genomic prediction of ordinal traits, studying factors that influence the performance of the models, and if there is a difference in the ranking of individuals. Genotypic and phenotypic information from a simulated population of 4,100 animals, genotyped by 10k markers (QTL-MAS Workshop) were used. 3,000 animals were used for estimation of the predictive ability and bias accessed through 5-fold cross-validation with five repetitions. The other animals were used as a population of selection. One ANOVA and the Ryan-Einot-Gabriel-Welch test were performed to verify, respectively, which factors influence significantly the genomic prediction and if there is a statistical difference between the models. The results show that the four main factors significantly (p < 0.05) affect the predictive ability of GEBVs (genomic estimated breeding values), and that heritability and the number of categories are the most influential factors. Only for ordinal trait 2, with a density of 9k, significant differences (p < 0.05) were observed between the predictive ability of the methods. In general, the BayesCπ method proved to be more efficient in the identification of relevant SNPs and in the ranking of individuals. Finally, there is a slight superiority of the BayesCπ method for the genomic prediction of ordinal traits.

Highlights

Methods for genomic prediction have been extensively evaluated for continuous traits in real (Croiseau et al, 2012, Resende Junior et al, 2012) and simulated (Atefi et al, 2016) data
The significance of the influence of factors, heritability, number of SNPs, model and number of categories, on the prediction of genomic estimated breeding values (GEBV) was evaluated by means of an analysis of variance (ANOVA) (Table 1)
Resende Junior et al (2012), comparing several methodologies regarding the accuracy of genomic prediction for 17 pine traits, corroborated with the results presented in this study, showing that for 11 of these traits, the BLASSO and BayesCπ methods presented the same predictive ability

Summary

Introduction

Methods for genomic prediction have been extensively evaluated for continuous traits in real (Croiseau et al, 2012, Resende Junior et al, 2012) and simulated (Atefi et al, 2016) data. It is in the interest of plants and animals breeding to determine the genetic variants associated with production, disease or resistance traits, which are not of a continuous and normally distributed nature (González-Recio & Forni, 2011). It is known that several traits of economic interest and importance in animal and plant production are measured in an ordinal scale. Ordinal scores are observed for disease resistance or susceptibility. Montesinos-López et al (2015) used genomic wide selection techniques to predict resistance to gray leaf spot, measured on an ordinal scale, in maize lines, considering three environments

Objectives

Methods

Results

Conclusion