Multivariate selection of AFLP markers associated with β-carotene in sweetpotatoes

A population that consisted of F₁ half-sibs and their parents was replicated in field plantings to generate roots to identify AFLP molecular markers and compare statistical models for selection of trait-linked markers using β-carotene content as the grouping variable. Genotypes were grouped into high and low β-carotene classes based on hierarchical cluster analysis. Association parameters between the high and low phenotypic classes of β-carotene and molecular marker profiles were made using logistic regression and discriminant analysis. Logistic regression selected 8 markers less that were associated with β-carotene content compared to discriminant analysis. Further analysis showed that logistic regression achieved 100% correct classification with the 9 markers selected compared to 17 markers that were required by discriminant analysis thus making logistic regression more efficient than discriminant analysis. Logistic regression based on Frequentist hypothesis testing differs from Bayesian based discriminant analysis. This difference may account for the percent correct classification observed.