83 Genotype-by-environment interaction with high-dimensional environmental data: an example in pigs

Abstract This study investigated using environmental covariates to model correlated herd effects and genotype-by-environment interaction (GE), as well as their impact on the prediction accuracy of genomic evaluations in pigs. We used air, dew/frost, wet-bulb, and earth-skin temperatures, relative humidity, rainfall, wind speed and direction. Average daily gain (ADG) and backfat thickness (BFT) from a terminal pig line were analyzed by 6 different models: MG, which included herd as an uncorrelated random effect; ME30, which considered 30 d of daily weather information to correlate herd effects; and ME100, which used 100 d of daily climate covariates; MGE30, same as ME30 but also considering the GE as the Hadamard product of the genomic relationship matrix and the environmental covariates matrix; MGE100, same as ME100 but including the GE; and MTM, a traditional multiple-trait model with each herd as a different trait. All animals were genotyped and phenotyped; therefore, the statistical model of choice was GBLUP. Validation was based on LR method, and focal animals were those born in the last year. Accuracy, bias, and dispersion did not change among models, except for MTM. Comparing all models against MTM, the accuracy increased by 38% for BFT, whereas only slightly increased for ADG. Among all 11 herds, only 4 showed a genetic correlation stronger than 0.80 for BFT, while this number was much greater for ADG. When the model has a variable accounting for environmental effects like herd in the case of MG, using weather information to correlate environments or adding the GE term has little benefit for genomic predictions.