Improvements in Prediction Performance of Ensemble Approaches for Genomic Prediction in Crop Breeding

Abstract

The refinement of prediction accuracy in genomic prediction is a key factor in accelerating genetic gain for crop breeding. The mainstream strategy for improving prediction performance has been to develop an individual genomic prediction model that performs better than others. However, this approach has limitations due to the absence of consistent outperformance by an individual genomic prediction model. This problem is aligned with the implication from the No Free Lunch Theorem, which states that the performance of an individual prediction model is expected to be equivalent to the others when averaged across all prediction scenarios. Hence, we applied an alternative method, combining multiple individual genomic prediction models into an ensemble, to investigate the potential for increased prediction performance. We developed a na&iumlve ensemble-average model, which averages the predicted phenotypes of individual genomic prediction models with equal weight. The prediction performance of the genomic prediction models was evaluated over two traits that influence crop yield, days to anthesis and tiller number, in a publicly available Teosinte Nested Association Mapping (TeoNAM) dataset. Here we show an improvement in prediction accuracy for the ensemble approach compared with the six individual genomic prediction models in more than 95% of prediction scenarios. The advantage of the ensemble was derived from a more comprehensive view of the genomic architecture of these complex traits provided by contrasting genomic marker effects from the individual genomic prediction models. These results indicate the potential for ensemble approaches to enhance the performance of genomic prediction and therefore genetic gain, in crop breeding programs.