Table 1. Validation results and most influential compounds of pairwise PLS-DA models from the log-transformed concentration variables with respect to the three visits.

Visit	Q2 and R2 values of thefirst 3 componentsa	NMC at 5 componentsand permutation testb	Involved compoundsc
V1 → V2	0.526, 0.663, 0.679/0.534, 0.676, 0.700	89.3 (84.0–93.8)/662 (625–695)	Unknown, 0.62 ppm↑; Unknown, 0.78 ppm↑; Alanine↑; Glycine↑; Lactose↑
V2 → V3	0.791, 0.868, 0.883/0.794, 0.873, 0.891	9.0 (8.0–10.0)/607 (581–633)	Unknown, 0.62 ppm↓; Unknown, 0.78 ppm↓; Alanine↓; Glycine↓; Lactose↑
V1 → V3	0.690, 0.782, 0.803/0.695, 0.788, 0.815	35.5 (32.3–39.0)/601 (578–627)	Unknown, 0.62 ppm↓; Unknown, 0.78 ppm↓; Alanine↓; Glycine↓; Lactose↑

^aA high ratio between Q² and R² confirms the validity of the models.

^bThe number of misclassifications (NMC) of the PLS-DA models relative to the random result from permutation testing serves as a performance estimate; 95% CI of the estimates in parentheses.

^cSpectral signals that could not be assigned to known metabolites are referred to as “Unknown”, along with the locations of their NMR signals. Arrows denote relative increase or decrease between visits. See also Table 2 and Fig. 2.