Experimental design and subject details
Method details
Quantification and statistical analysis
Utilization of primer sets derived microbiome data in the statistical analysis
Converting OBITools intermediate fasta files to QIIME ready format
Clustering of microbial marker gene amplicon sequences and picking representative de novo species OTU
Assigning taxonomy to OTU
Creation of OTU tables and sample subsetting and subsampling
Correlating microbial domain cell count
Correlating microbial domain cell counts to experimental variables
Correlating microbial domain richness to experimental variables
Prediction of phenotypes and other experimental variables by core microbiome
Prediction of phenotypes by core microbiome while correcting for diet
Prediction of phenotypes by diet components
Bovine genotype quality control
Testing association of the global rumen prokaryotic core with host genetics
Creation of genetic relationship matrix
Heritability estimation
Heritability confidence interval estimation
Bovine genome SNPs—Microbe association effort
Estimating kinship matrix
Genomic prediction
Associating microbes’ abundance with experimental variables
Inference of microbial interaction network within domains
Inference of interdomain microbial network
Comparing phylogenetic relatedness of core prokaryotic microbes to random sampling
Examining core and trait-related microbiome for taxonomic enrichment
Comparing heritable microbes to other core miocrobes’ ability to explain experimental variables
Seasonality test
As an additional analysis to further verify our findings of core microbiome explainability (by prediction) of host phenotypes and experimental variables, we repeated that analysis using RF regression.
The abundances of the core microbes within each farm were used as features fed into a RF regression model (21, 22) to predict each of the traits (separately). Our approach followed a leave-one-out cross-validation methodology where, in each iteration, one sample (animal) was omitted from the entire set, and the model built from all the other animals (training set) was used to predict the trait value of the excluded sample (animal). Thereafter, the prediction R2 value between vector of actual and predicted values was calculated using R CARET package function R2.