Different feed efficiency modeling approaches for the prediction of genomic breeding values in lactating dairy cows

Feed is the main cost of production in dairy farming. Any improvement in feed efficiency (FE) would increase marginal profit and sustainability and mitigate the environmental impact of dairy farming. In this study, we applied single-step genomic best linear unbiased prediction to different feed-efficiency metrics using records collected from Nordic Red dairy cattle (RDC). The main objective was to compare different metrics in terms of their effectiveness in selecting more feed-efficient animals. Weekly observations (n = 22,071) of dry-matter intake records from 791 RDC cows collected from 1998 to 2021 were used in this study. The pedigree consisted of 5,604 individuals, of which 1,489 animals were genotyped. Different modeling approaches, including conventional residual feed intake (RFI), regression on expected feed intake (ReFI), two multi-trait residual feed efficiency indices (RFI_Index and RZFE), and energy conversion efficiency (ECE) were analyzed. For the ReFI approach, two alternatives for predicting the expected feed intake, namely, a prediction equation tailored to the RDC data and a prediction equation based on Holstein dairy cow data proposed by the National Academies of Sciences, Engineering, and Medicine (NRC 2021), were compared. First, a BLUP model was developed, and the necessary variance components were estimated for each approach. Then, pedigree-based and genomic-enhanced breeding values (PEBV and GEBV, respectively) were estimated using either reduced or full datasets. For model validation, PEBV and GEBV estimated using the full dataset were regressed on PEBV and GEBV estimated using the reduced dataset, respectively, to measure bias, dispersion, and prediction accuracy (PAC). The heritability estimates of different residual metrics ranged from 0.23 for RFI to 0.30 for ReFI_NRC2021, and the repeatability estimates ranged from 0.48 to 0.52. The estimated heritability and repeatability of ECE were 0.23 and 0.56, respectively. For all metrics, the use of genomic information increased PAC. However, there were discrepancies between the metrics in terms of the magnitude of PAC, with the PAC being the highest for ReFI_RDC and the lowest for RFI_Index. Similarly, ReFI_RDC had the lowest bias, while the highest bias was estimated for RFI_Index. In addition, RZFE and ReFI_RDC showed lower dispersion. The correlations between GEBV of the residual metrics and the GEBV of ECE were lowest for RFI_NRC2021 and RFI and highest for ReFI_RDC. Among the metrics compared, ReFI_RDC and RFI_Index showed the highest effectiveness in selecting efficient cows. This indicates that the use of appropriate partial regression coefficients and the type of modeling are vital in breeding programs aimed at enhancing FE.