Fig. 2

a Principal component analysis (PCA) of untargeted metabolite data of pasture-finished (Bi_pasture) and pen-finished (Bi_pen) bison displayed clear and distinct separation between the pasture-finished and the pen-finished samples along component 1 (explaining 20.74% of the variation). b Random forest analysis based on the biochemicals detected in this dataset resulted in a predictive accuracy of 100% between dietary conditions. This value is greater than random chance (50%), suggesting these metabolites are of interest as biomarkers. c Random forest analyses obtained from grass-fed and grain-fed bison samples. The y-axis represents the biochemicals in order of importance for group classification, from top to bottom. The legend represents the type of biochemical and the table next to the chart shows the prediction accuracy based on the random forest results (actual against predicted group)