Fig. 2

Ensemble clustergrams by subject (top) and principal component analysis (PCA) loading plots (bottom). A Pre-season, B In-season, C Full study, and D Seasonal difference (in—pre). Ensemble clustering was performed in Python for pre-season, in-season, both seasons (“full study”), and seasonal difference (in-season – pre-season) of biomarker levels and rancher subject IDs using the single, complete, average, and ward linkage functions with the Euclidean distance, and the single, complete, and average linkage functions with the squared-Euclidean, cosine and Chebyshev distance. PCA biplots were generated in MATLAB to identify biomarkers with relatively large leverage within the dataset (“distinct”) for profiling and post-hoc statistics. In-season biomarker responses were more distinct than pre-season, with rancher subjects clustering together in more unique subclusters during calving season. MMP-3 and leptin were distinct in all PCAs, while C2C was only distinct during pre-season