Application of support vector regression for prediction of Grey Leaf Spot resistance using high density molecular data.

ORNELLA, LEONARDO; PEREZ, PAULINO; TAPIA, ELIZABETH; CROSSA, JOSÉ

Maize genetics cooperation news letter

University of Missouri, Columbia

Lugar: Madison, Wisconsin; Año: 2012 vol. 86 p. 1 - 3

1090-4573

The causal organism associated with the disease is Cercospora zeae-maydis. Pandemic in Africa, GLS is now recognized as one of the most significant yield-limiting diseases of maize worldwide (Ward et al., Plant Sci. 83: 884-895). If multigenic, the nature of a resistance impedes the use of MAS (marked assisted selection) for the generation of reliable, resistant genotypes. A valuable alternative is Genomic selection (GS), because GS estimates breeding values of individuals using the sum of all marker effects, many limitations associated with detecting significant marker trait associations are bypassed and the total genetic variation for the traits of interest can be better captured by the markers (Heffner et al., Cro Sci. 50:1681-1690) . Prediction of genetic values can be carried out using parametric or nonparametric approaches. Parametric models, such as BayesA (Meuwissen et al, Genetics 157: 1819-1829) , Bayesian Lasso (Crossa et al., Genetics 186: 713-724) or Ridge Regression (Piepho, Crop Sci. 49: 1165-1176) are the most commonly used. However, they are not flexible enough to incorporate complex gene action (e.g., dominance or epistasis). Support vector machines are considered a state-of-the-art machine learning algorithms for classification and regression (SVR). Due to their theoretical foundations, SVR is very suitable for GS applications. Besides, unlike parametric models, no strict assumption is made regarding the form of the genotype–phenotype relationship. Rather, this relationship is driven primarily by the data.