Parsimonious Gene Correlation Network Analysis (PGCNA): a tool to define modular gene co-expression for refined molecular stratification in cancer

Radical edge reduction enhances the resolution of biology in gene co-expression modules. a Enrichment of gene ontology and signatures was assessed using a scaled cluster enrichment score (SCES) and compared between data clustering generated with FastUnfold, Hierarchical clustering or k-means clustering using either the total correlation data (All) or parsimonious matrices with edges per gene (EPG) thresholds between 3 and 10. Violin plots display the distribution along with median (blue square) and the IQR. b, c density plots of the module size (gene-number; x-axis) vs module SCES (Scaled Cluster Enrichment Score; see methods; y-axis) across the 100 best clustering for EPG3 (Edge Per Gene 3) or the WGCNA sigmoid adjacency function varying the shift (μ, 0.1, 0.5 and 0.9). b BRCA data and c CRC data. Beneath the graph the results are displayed in tabular form, showing the median module number, median modular number with gene membership >5, edge number retained after filtering (edges <0.01 removed), median sum of module SCES, median module SCES, and the percentage of connected genes (after filtering; edges <0.01 removed