Fig. 3

GNA benchmarking. a Performance of all GNA tests on SBM benchmark data. On the left column, we report the True Positive Rate (TPR) and False Positive Rate (FPR) for \(U_H\), while on the right column we report the same metrics for \(U_{SP}\). On the x-axis, we show different geneset sizes, while we denote the overlap between the tested genesets with colors. For example, for size 50 and \(4\%\) of overlap the two geneset share 2 nodes. We notice that \(U_{H}\) has TPR \(>0.95\), while \(U_{SP}\) is consistently below 0.75. Moreover, the FPR analysis confirms better performance for \(U_H\), albeit it is skewed by many overlapping nodes. On this point, when two genesets share 6 or more nodes out of 50, \(U_H\) always considers them as positives