Fig. 15

Blood-based biomarkers for tissue/organ-specific rhythmicity. Performance of one-and two-sample models derived from the training set when used to predict timing of ANT (antrum) and PVN (paraventricular nuclei) rhythmicity in the validation set. Predicted timing of ANT vs. sample clock time for each sample in the validation set for a the one-sample (n = 68) and b two-sample (n = 38) PLSR model. c Proportion of predictions of timing of ANT vs. cumulative absolute error for the one- and two-sample PLSR model. Predicted timing of PVN vs. sample clock time for each sample in the validation set for d the one-sample (n = 68) and e two-sample (n = 38) PLSR model. f Proportion of predictions of timing of PVN vs. cumulative absolute error for the one- and two-sample PLSR model. The cumulative frequency distribution of absolute error can be used to identify the proportion of samples that can be predicted within a given error range, where the highest proportion combined with smallest error range denotes the better prediction model