Metabolomic-genomic prediction realizes small increases in accuracy of estimated breeding values for daily gain in pigs

Abstract

Abstract Background Metabolomic profiling of blood samples can be done on selection candidates and could be a valuable information source for genetic evaluation of pigs. We hypothesized that integrating metabolomic data from pigs without individual phenotypes into the metabolomic-genomic best linear unbiased prediction (MGBLUP) model would generate estimated breeding values (EBVs) with a higher accuracy compared to what would be obtained without metabolomic data. We tested this hypothesis by predicting breeding values for average daily gain (ADG) using phenotypic, genomic, and metabolomic data. MGBLUP models were fitted to average daily gain of 8174 Duroc pigs that were genotyped and profiled for metabolomic features. Approximately half the pigs were males from a test station and the other half were females from breeding herds. Variance components were estimated, and we employed two validation schemes: test station to breeding herd validation and fivefold cross-validation. Accuracies of EBVs in the validation population were computed by combining results on predictive abilities with results on increases in accuracies from the linear regression method. Results Parameter estimates from MGBLUP showed a direct heritability of ADG of 0.15, a proportion of variance explained by metabolomic features of 0.18, and a heritability of metabolomic intensities of 0.14, together resulting in a total heritability of 0.17. Thus, the majority of the heritability was not mediated by the metabolome. For the test station to breeding herd validation, the accuracies of EBVs were 0.60 for genomic best linear unbiased prediction (GBLUP) with genotypes in validation population, 0.61 for MGBLUP with genotypes in validation population, 0.62 for MGBLUP with genotypes and metabolomic features in validation population, 0.72 for GBLUP with genotypes and phenotypes in validation population, and 0.74 for MGBLUP with genotypes, phenotypes and metabolomic features in validation population, whereas the corresponding numbers were 0.87, 0.87, 0.87, 0.91 and 0.92 for the fivefold cross-validation. Therefore, small increases in accuracies were observed when including metabolomic features. Conclusions The inclusion of metabolomics data provided small improvements in the accuracy of genetic evaluations for average daily gain in pigs. Further work will be needed to investigate, e.g., alternative time points for blood sampling, metabolomics on samples of other tissues, and other traits.