Fig. 4

A PCA and (B) PLS-DA scores plots indicating the distribution of samples (healthy controls and all untreated patient samples, n = 53). Groups mainly overlapped at (A) but were more homogenous and could be better distinguished from each other at (B) mainly because of changes linked to inflammation and the breakdown of molecules as fuel during chronic inflammatory diseases such as this. While the PLS-DA model (B) did not perform optimally in cross-validation for a one-component model (accuracy = 0.54), it did validate during permutation testing (p-value =  < 0.05)