Fig. 2

The hiPCA can stratify healthy and unhealthy groups. a The balanced accuracy of hiPCA under various H+/H− thresholds on discovery data (PVE 0.9 and confidence rate 0.9). b The balanced accuracy of hiPCA under various H+/H− thresholds on validation data (PVE 0.9 and confidence rate 0.9). c The respective features are selected under various H+/H− thresholds. d The intersection between GMHI 50 features and KS 92 features. e The surf plot of balanced accuracy by 92 features under different parameter configurations. The yellow zone represents the balanced accuracy results from the discovery set, while the red zone represents the balanced accuracy results from the validation set. f Stacked accuracy bar over different weighting factors from discovery data and test data with GMHI 50 features. g Stacked accuracy bar over different weighting factors from discovery data and test data with KS 92 features. TP and TN denote the true positive rate and true negative rate, respectively, and the balanced accuracy is the average between TN and TP in each single bar. h,i Box plot of hiPCA (PR-50 and KS-92) in healthy and nonhealthy groups. j,k Box plot of hiPCA (PR-50 and KS-92) in healthy and 12 nonhealthy phenotypes. l The hiPCA (PR-50 and KS-92) accuracy rates over different phenotypes. All P-values shown above the box plots were found using the two-sided Mann-Whitney U test: *, P ≤0.05; **, P ≤0.01; ***, P ≤0.001; ns, not significant. The sample size of each group is shown within parentheses