Fig. 4: DynamicGP outperforms the baseline models on the maize dataset.
a, The prediction of the complete time-series in iterative and recursive configurations in unseen lines. b, The predictions of only the last five timepoints in unseen lines in models trained with the data from the first 20 timepoints in the maize dataset. We determined the difference in performance of the iterative and recursive versions of dynamicGP and baseline RR-BLUP models, for traits whose predictability is above a threshold value, specified on the x axis. The number of traits with predictability larger than a specified threshold value is indicated above the respective bar. The best predicted traits with dynamicGP exhibited the greatest difference from the baseline predictions. The bars indicate the mean (±standard deviation) difference in trait prediction accuracies of traits that have mean prediction accuracies above the threshold. The number of samples (n) is specified on the top of each bar.
