Fig. 2

Global expression correlation attenuation and extremely low hub retention of pathways. a breakdown of differentially co-expressed gene links (DCLs). Each DCL is characterized with a pair of correlation values corresponding to the two comparator conditions, respectively, and DCLs are categorized into four types on account of the signs and changing trend of the paired correlation values. Diff signed, DCLs of two extreme correlation values in opposite signs. Same signed negative, DCLs of two negative correlation values. Increased positive, DCLs showing correlation increment toward extreme positive values. Decreased positive, DCLs showing correlation decrease from extreme positive values. b breakdown of pathways by predominant correlation change direction. Dissolved, more gene pairs have decreased correlation. Consolidated, more gene pairs have increased correlation. Maintained, even share of gene pairs with increased correlation and gene pairs with decreased correlation. c one hundred times of random permutation of patients’ class labels were performed and GSNCA was implemented on the permutated datasets, with respect to all 369 covered pathways. The real hub constancy rate (3/27) and hub retention rate (1/44) was compared against the empirical distributions resulting from permutations. d hub constancy rates and hub retention rates in real data analysis (red line) and permuted analyses (grey histogram), where one hundred times of random permutation of patients’ pathway annotations preceded GSNCA running. Technically, permuting patients’ pathway annotations was equivalent to shuffling the gene-to-pathway mapping relations, thus achieving random organization of genes to meaningless pseudo-pathways while maintaining the same pathway size profile. The real hub constancy rate (3/27) and hub retention rate (1/44) was compared against the empirical distributions resulting from permutations