R-squared between predicted and observed values () estimated using different number of SNPs (where K = 1,000), models, and validation designs.

BL = Bayesian LASSO, GH = Goddard-Hayes, and GY = Yang study (see Materials and Methods for elucidation). 110-fold cross validation, where the training set comprised 4,605–4,606 individuals. 2Models were trained using the original cohort (N = 1,493) and predictive ability was assessed in the Offspring cohort (N = 3,624). 3Data was assigned at random to a training set (N = 1,493) and predictive ability was evaluated in the remaining individuals (N = 3,624). This was repeated 10 times; each time individuals were randomly assigned into training/testing sets. Results are averaged across the ten replicates.