Prediction of maize hybrid silage performance using marker data:: Comparison of several models for specific combining ability

Methods that could predict F-1 maize (Zea mays L.) hybrid performance with some accuracy prior to field evaluation are of particular interest, This study was conducted to evaluate the efficiencies of the conventional additive model based on the general combining ability (GCA) as a control and of three different predictions of the specific combining ability (SCA), The first approach was based on the hypothesis that the degree of SCA expressed by a single-cross is related to the marker distance between its parental lines. The second approach was based on a factorial regression model of interaction, where markers were used by means of principal component analysis to generate covariates for SCA, The third approach was adapted from the best linear unbiased prediction (BLUP) of SCA, where covariances between hybrid SCAs were estimated with marker data, Efficiencies were evaluated by means of a cross-validation procedure for silage performance of a 21 by 21 half-diallel population among maize inbreds, This procedure was applied to (i) all hybrids and (ii) hybrids between unrelated parents only. In situation (i), introducing a distance term in the model accounted for up to 73.6% of the variation in the hybrid performance observed, whereas the corresponding efficiency of the GCA model was 63.4%, The introduction of a distance term did not modify prediction efficiency in situation (ii) whereas the use of the factorial regression model or the BLUP approach led to moderate improvements, These results suggested that efficient approaches can be proposed to predict hybrid silage yield and that prediction of SCA is mostly justified in situations where coancestry among inbreds is unknown or only suspected.