Fig. 4.

Decision rules for identifying extreme and non-extreme patient groups that take into account the specific shape of the gene expression distribution. a For symmetric distributions, the extreme patient group is represented by the samples falling in either the upper or lower percentile of the distribution as shown by the purple tails (in our analysis, the tenth percentile is used). The non-extreme patient group corresponds to the samples falling between these two percentile cut-offs as shown by the green region. b. For asymmetric distributions, the extreme patient group corresponds to samples only in the first or last percentile depending on the shape of the asymmetry, as shown by the one-sided purple tail. The remaining region of the distribution represents the non-extreme patient group. c For bimodal distributions, the split is determined by a clustering algorithm applied to the expression data to identify which patients belong to one group (mode) versus another. For genes in the bimodal expression category, the definition of extreme and non-extreme patient groups is not relevant, and instead we identify two patient groups for comparison, as shown by the purple and green regions. d Theoretical example of two survival curves constructed for patients in Groups 1 and 2 as defined in a, b or c