enrichment of gene sets in individual samples. Then we estimated the degree of association between each hallmark’s ssGSEA profile and the Hedgehog vs. photoreceptor activation phenotypic distinction using the Information Coefficient (IC, see Methods). In fact, any standard method for differential analysis could be used for this purpose. An empirical sample-label permutation test (500,000 random permutations) was used to estimate the statistical significance of the scores (i.e., the p-value and False Discovery Rates (FDR) in Figure 1A). Notably, if we rank the hallmarks according to their association with the Hedgehog versus photoreceptor phenotypes then the top-scoring hallmark, i.e., the one with the highest IC value (IC: 0.7346, p-value: 2×105, FDR:1×105), is indeed the Hedgehog hallmark, consistent with prior findings in the literature (Cho et al., 2011). A heat map of the results along with the hallmarks ranked by IC appears in Figure 1A. Thus we see the sensitivity of the Hedgehog hallmark in detecting its relevant biological process i.e., oncogenic activation of the sonic hedgehog pathway in the context of a disease subtype.
