Fig. 3: Consistency of trait heritability across time influences the prediction accuracy of dynamicGP on the maize dataset.
a, The longitudinal prediction accuracy of representative traits corresponding to the maximum, minimum and mean, as well as the first and third interquartiles of all 50 traits across the 24 predicted timepoints (Table 1). The operator Ar was obtained from dynamicGP using 5-fold cross-validation (CV). The solid and dashed lines represent predictions from the iterative and recursive versions of dynamicGP, respectively. b, The time-resolved, scaled mean trait values across all accessions, depicting the different dynamics of traits. c, The heritability of the five representative traits across time (see Extended Data Fig. 1 for depiction of the dynamics of these traits). The colours for the particular traits are maintained in a–c. d, A Pearson correlation between the mean prediction accuracy of traits from iterative dynamicGP and the coefficient of variation of heritability estimates of traits across time (The Pvalue is derived from two-sided test from a sample of 50 points). e, A category membership of traits in the full dataset (498 image-drived traits) compared with the traits that were selected by our clustering method for inclusion in the analysis (50 image-derived traits).
