Mean correlation between predicted and observed phenotypic values, and mean proportion of variance associated with models trained in test sets and evaluated in disjoint validation sets over ten folds of cross-validation of rare allele scan QTL models.

In each fold, the forward selection algorithm for the rare allele scan QTL was performed in a training set of 90% of the population individuals and the resulting model was used to predict the trait values of the held-out 10% validation set of individuals. The correlation between predicted and observed values was estimated from the null model containing only environmental covariates but no genetic effects (r_null), a model adding principal components of the genome-wide marker data (r_pc), a model adding selected rare allele scan QTL effects (r_pc), and a model including both principal components and QTL (r_full). The change in correlation from adding QTL to the null model (r_qtl_vs_null) and the change in correlation from adding QTL to the PC model (r_pc_vs_qtl) was measured for each trait. Similar values are estimated for the proportion of trait variation explained (R2) in the validation set for each model.
(CSV)