Multi-omics prediction of swine production traits with linear mixed models

Background Current prediction of phenotypes often relies on single-nucleotide polymorphism (SNP) data from DNA microarrays. Advances in biotechnology now enable large-scale collection of complementary omics data, such as transcriptomics and epigenomics. How much these additional molecular layers improve prediction accuracy, compared to SNP-based models, merits investigation. We evaluated whether using RNA transcript abundance and DNA methylation levels from loin muscle biopsies increased phenotype prediction accuracies for four traits in swine: average daily gain (ADG), backfat thickness (BFT), residual feed intake (RFI), and strength score (STR). We obtained transcriptomic (RNA sequencing) and epigenomic (reduced representation bisulphite sequencing) data of 443 boars representing the Duroc, Landrace, and Yorkshire breeds. We fitted linear mixed models incorporating genomic, transcriptomic, and epigenomic effects, and evaluated predictive performance by the correlation between preadjusted and predicted phenotypes obtained through cross-validation. Results When animals from all breeds were used, the predictive accuracy (measured by the correlation between predicted values and preadjusted phenotypes in leave-one-out cross-validation) for ADG ranged from 0.12 with genomic best linear unbiased prediction (GBLUP) to 0.37 with transcriptomic–methylomic BLUP (TMBLUP) and genomic–transcriptomic–methylomic BLUP (GTMBLUP). For BFT, accuracies ranged from 0.18 with methylomic BLUP (MBLUP) to 0.39 with genomic–transcriptomic BLUP (GTBLUP) and GTMBLUP. For RFI, accuracies ranged from 0.19 (pedigree BLUP, GBLUP, MBLUP) to 0.58 (GTBLUP). For STR, accuracies ranged from −0.27 (MBLUP) to 0.11 with transcriptomic BLUP (TBLUP), TMBLUP, GTBLUP, and GTMBLUP. Genomic–methylomic BLUP (GMBLUP) models exhibited intermediate predictive performances. Conclusions Incorporating transcriptomic and epigenomic data from loin muscle biopsies increased the accuracy of phenotype prediction relative to standard GBLUP, with transcriptomic information serving as the primary driver of these gains. Among the single-omics models, TBLUP attained the highest accuracies, while MBLUP exhibited the lowest performance. The combined GTMBLUP model was among the best-performing models overall, although gains relative to GTBLUP or TMBLUP were modest when present.