Fig. 5.
Prediction accuracy (ρ) in landrace KE for per se performance (PP) in the DH and GC lines as a function of sample size N (A), for prediction of PP and testcross performance (TP) at the maximum available number of lines (Nmax; B), for predictions within and across populations for PP (C) and TP (D) in DH and GC, and for across-landrace prediction for PP from KE (training on PE; E). Traits are plant height at V6 stage (PH_V6), final plant height (PH_final), and flowering time (FF) in PP and TP and dry matter content (DMC) and total dry matter yield (TDMY) in TP. For each N (A), sampling of lines was repeated 100 times, and 10 times fivefold cross-validation was carried out within each sample, yielding the basis for calculating the presented means and 95% quantiles (shaded areas around the curve). Prediction across and within populations as well as across landraces was carried out by randomly sampling N = 200 and N = 75 lines for training in PP (C and E) and TP (D), respectively, for predicting N = 50 (PP; C and E) or N = 25 (D) genotypes of the same or corresponding population (C and D) or the same population of the other landrace (E). Sampling was repeated 100 times. The violin plots (C–E) show all 100 values, with the diamonds indicating the means. Black dots show values of the prediction accuracy estimated from models where the genomic variance estimate was not significant (likelihood-ratio-test, P > 0.05).