A Stacking Ensemble Learning Framework for Genomic Prediction

Machine learning (ML) is perhaps the most useful tool for interpretation of large genomic datasets. However, performance a single machine method in selection (GS) currently unsatisfactory. To improve predictions, we constructed stacking ensemble framework (SELF), integrating three methods, to predict estimated breeding values (GEBVs). The present study evaluated prediction ability SELF by analyzing real datasets, with different genetic architecture; comparing accuracy SELF, base learners, best linear unbiased (GBLUP) and BayesB. For each trait, performed better than which included support vector regression (SVR), kernel ridge (KRR) elastic net (ENET). was, on average, 7.70% higher GBLUP Except milk fat percentage (MFP) traits, German Holstein dairy cattle dataset, was more robust BayesB all remaining traits. Therefore, believed that SEFL has potential be promoted estimate GEBVs other animals plants.