Genetically regulated omics integration improves genomic prediction of complex traits in pigs

Regulatory interactions across biological layers are highly complex. Intermediate omics (e.g., transcriptomes) can reveal how DNA variation influences traits, but raw omics measurements are noisy and commonly mix genetic and non-genetic signals. We developed the Genetically Regulated Additive and Dominance model (GRAD), a single-step framework. GRAD first decomposes each omics feature into genetically regulated additive and dominance components, and then jointly models additive and dominance effects at both the genomic and intermediate-omics levels to predict complex traits. We evaluated GRAD using simulations with dominance contributions ranging from 0 to 50% of genetic variance, and using a Landrace pig dataset (1 494 genotyped individuals, of which 105 had whole-blood transcriptome sequencing). In simulations, GRAD increased up to 21.98% (4.73% on average) accuracy gain relative to standard genomic best linear unbiased prediction, and it also outperformed models that used raw omics data or only additive omics signals. In the pig data, GRAD improved prediction for age to 100 kg and total number of piglets born by 7.4 and 14.9%, respectively. Our findings demonstrate that explicitly extracting genetically regulated additive and dominance signals from omics data increases the genetic information available for prediction. However, practical benefit depends strongly on the proportion of animals profiled and on sampling tissues that are mechanistically relevant to the target trait.