Fig. 3.

Difference in accuracies of genomic predictions between the supervised and self-training algorithms. Accuracy of genomic prediction was tested for 252 animals in the testing set, and measured as the a correlation or b mean squared error between predicted and measured residual feed intake phenotypes. The number of initial reference animals with measured phenotypes was 300, 400, or 500, and a total of 200, 400, 600, and 800 animals with self-trained phenotypes were added to the training set using the self-training algorithm. Error bars indicate the 95% confidence intervals for each mean of 100 replicates