Fig. 3

An ensemble method to depict the rumen microbiome signature of residual feed intake (RFI) in lactating dairy cows. The method consisted of summarizing the differential abundance analyses of some of the most robust methods available in the literature. 3.A. Distribution of RFI means among dairy cows in the studied cohort depicting differences between US and Canadian cows. 3.B. Shannon index, as a representation of alpha-diversity differences between the extremes of least and most efficient cows of the RFI phenotype, showing that the rumen microbial community diversity is lower among the most efficient cows. 3.C. Principal coordinate analysis (PCoA) of the rumen microbiome showing that the extremes of least and most efficient cows of the RFI phenotype in the study have a set of different microorganisms that could potentially modulate this trait. 3.D. Summary of all statistically significant microbial taxa (P < 0.05) in the RFI phenotype that have greater potential to not be affected by the sources of variation evaluated in the study. Ensemble analysis was performed using a STEPWISE selection (lowest AICC) in linear regression analysis (microbial taxa tested as microbial count, relative abundance, and centered-log ratio), machine learning regression (microbial taxa tested as microbial count, relative abundance, and centered-log ratio), and four independent differential abundance analyses tests: ALDEx2, ANCOM-BC, MAaslin2, and LinDA (contrast of phenotypic extremes; least efficient n = 50 cows vs. most efficient n = 50 cows). Microorganisms are ranked according to their significant prevalence across tests and represent potential key microbial taxa contributing to the trait of interest. 3.E. Correlation between normalized (CLR) key microbial players and RFI in the extreme population of least and most efficient cows. Only microbial taxa different in at least 4 statistical tests are displayed in these results