Fig. 5

MixEnrich shows higher performance than other single-subject and cohort-based methods (the latter utilized on small samples). Each boxplot corresponding to the N-of-1-pathways methods (MixEnrich in purple, MD in green, and Wilcoxon in orange) consists of 15 AUCs resulting from 15 tested patients. Each boxplot corresponding to the cohort-based methods (DESeq + Enrichment in red and GSEA in blue) includes 50 AUCs resulting from 50 distinct subsets of the 15 tested patients (Validation case study of head and neck cell carcinoma patients). Cohort-based methods were performed across 3, 6 and 12 patients (Pt). The number of distinct subjects is shown below the horizontal axis as human icons to further illustrate how many distinct subjects are required in cohort-based analyses to obtain improvements of the AUC (vertical axis). In addition, the three single-subject analyses predict between 200-300 candidate pathways at FDR = 1%, while cohort-based statistics operating on 3 to 12 individuals predict only 50 pathways at FDR = 5% and over 200 at FDR = 20% (data not shown), which explains in part the observed differences in accuracies