Metagenomic prediction of methane emissions in sheep using single- and multi-matrix BLUP models with taxonomic and functional microbial features
BackgroundEnteric methane emissions from ruminant livestock represent a major greenhouse gas contributor, yet identification of high- and low-emitting ruminants remains expensive and logistically challenging for agricultural methane mitigation strategies. Ruminal microbial profiles derived from long-read sequencing technology provide a potential proxy to predict methane production. The optimal bioinformatic pipelines for processing long-read metagenomic data to perform methane predictions have yet to be determined. Here we evaluated how different metagenomic analysis pipelines affect methane predictive model accuracy in grazing sheep. ResultsWe applied three bioinformatic pipelines to characterize the taxonomic and functional features of rumen microbiomes from 396 sheep. Functional abundance features were annotated from Clusters of Orthologous Genes (COG) or Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. The single-matrix model using COG features achieved the highest microbiability (m2 = 0.942: proportion of variance component explained by microbial features) and predictive accuracy (5-fold cross validation r = 0.609: Pearsons correlation between predicted and observed values). Both functional features outperformed all taxonomic features across all three pipelines in predictive accuracy. The multi-matrix models combined functional and taxonomic features slightly improved methane predictive accuracy across both 5-fold cross-validation and leave-one-day-out validation compared to the models using functional features alone. ConclusionsThese findings demonstrate the potential advantages of using long-read metagenomic data to predict enteric methane emissions in ruminants. COG-based functional features achieved the highest predictive accuracy among all feature types, suggesting that functional annotation of existing long-read sequences is sufficient for accurate methane prediction without requiring complementary taxonomic data.