Fig. 6. Detailed results on the stability and usability performances of the DR methods.

a The stability of the DR methods. We calculated the stability of the DR methods, across different bootstrap datasets generated from the Oetjen cohort’s Sample A. b The impact of parameter tuning on the accuracy of the DR methods, in comparison with the default settings. In the left panel, we showed the optimal settings’ (among all tested settings, Supplementary Data 4) increases in ranks (over default settings’ ranks) divided by the total number of configurations for each method and each category. The right panel shows the relative performance of the default settings (with respect to the optimal settings), quantified by the ranks of the default settings divided by the maximum ranks achieved by the optimal settings for each method. c Usability scoring for the DR methods. All detailed usability criteria were displayed for all DR methods, in the format of a heatmap. Teal shading represents the presence of a feature in the heatmap and better overall usability in the bar graph, whereas orange shading indicates the lack of a feature and poorer usability. Source data are provided as a Source Data file.