Improving genomic prediction accuracy for methane emission and feed efficiency in sheep: integrating rumen microbial PCA with host genomic variation using neural network GBLUP (NN-GBLUP)

Methane emissions from ruminant livestock pose a significant challenge to mitigating climate change. Genomic selection offers a promising approach to reduce methane emissions, but prediction accuracy remains low due to the high cost of measuring methane emissions. Integrating rumen microbiome composition (RMC) data may improve genomic prediction accuracy, yet the high dimensionality of RMC data presents computational challenges. This study aimed to (1) evaluate the effectiveness of principal component analysis (PCA) for reducing RMC data dimensionality while retaining essential information, and (2) assess whether incorporating PCA-reduced RMC data as intermediate traits in a Neural Network Genomic Best Linear Unbiased Prediction (NN-GBLUP) model improves genomic prediction accuracy for methane emissions and feed efficiency traits in sheep. For the first objective, Principal Components (PCs) explaining 100% of variation effectively captured RMC information, with microbiability estimates closely matching those from the full dataset. For the second objective, the NN-GBLUP model incorporating PCA-reduced RMC data improved prediction accuracy compared to standard GBLUP methods. Prediction accuracy for methane emissions increased from 0.09 to 0.30 in train-test validation and from 0.15 to 0.27 in five-fold cross-validation using PCA components explaining 25% of total RMC variation. For residual feed intake, accuracy improved from 0.25 to 0.37 in train-test validation and from 0.25 to 0.34 in cross-validation. Optimal PCA components varied by trait, with 25% and 50% components showing the best results. Prediction accuracy did not improve for carbon dioxide emissions, live weight, and mid-intake, indicating trait-dependent microbiome influence. Principal Component Analysis reduced the dimensionality of rumen microbiome data while preserving essential biological information. The integration of these PCA-reduced data with host genomic information through an NN-GBLUP model substantially improved genomic prediction accuracy for methane emissions and feed efficiency in sheep. Principal components explaining 25% and 50% of the variation yielded the highest accuracy, whereas higher components (75% and 95%) reduced accuracy for methane traits. This approach shows promise for implementing genomic selection strategies to reduce methane emissions and improve feed efficiency in ruminant livestock in a computationally efficient manner, thereby contributing to climate change mitigation efforts in agriculture.