Fig. 1: Identification of dysregulated and prognostic genes/pathways in ILCs.

a Schematic of the methodology, divided into 3 steps. Step 1 constitutes feature selection using the Metabric discovery cohort with mRNA abundance profiles of ILC (n = 148) and normal breast tissue samples (n = 144). ILC and normal mRNA abundance profiles were compared by performing differential gene expression and differential variance analysis. The resulting genes were subject to pathway over-representation analysis to identify dysregulated pathways in ILCs. To remove pathway redundancy, pathways with shared genes due to nested hierarchy were collapsed into pathway modules called a “cluster of pathways” (CP). In step 2, a survival model for each CP was created using Metabric mRNA abundance profiles and clinical outcome data. In step 3, each CP-based survival model was applied to the mRNA abundance profiles of the six validation cohorts separately to predict patient risk score. These predicted risk scores were dichotomised into risk groups using the discovery cohort’s median. The resulting risk groups from six validation cohorts were combined and subsequently correlated with patient outcome. b Venn diagram showing the overlap of differentially expressed and differentially variable genes between ILC (n = 148) and normal breast tissue samples (n = 144), identified using the Metabric dataset. c t-SNE clustering of tumour and normal samples using 1398 ILC-dysregulated genes.