Transcriptome-Based Prediction Of Hybrid Performance With Unbalanced Data From A Maize Breeding Programme

mRNA transcription profiles are an alternative to DNA markers for predicting hybrid performance. Our objective was to investigate their prediction accuracy in an unbalanced maize data set. We focused on the effectiveness of preselecting a core set of genes for transcription profiling and on the comparison of prediction models. A total of 254 hybrids were evaluated for grain yield and grain dry matter content. The mRNA transcripts of a core set of 2k genes and the genotype of 1k AFLP markers were assessed in the parental lines. Predictions based on transcriptome-based distances determined from the 2k core set of genes resulted in prediction accuracies below 0.5 and could not reach the high accuracies observed with a 46k micro-array in earlier studies. Predictions based on ridge regression resulted in prediction accuracies greater 0.6. Only marginal differences were observed in the prediction accuracies of mRNA transcripts compared with AFLPs. We conclude that mRNA transcription profiles are suitable for hybrid prediction with ridge-regression models in unbalanced designs, even if limited resources allow only transcription profiling of a core set of genes.