Fig. 2

Compilation and functional characterization of IPF prognostic predictor gene set. a A flowchart illustrates the procedures and approaches used for IPF prognostic predictor gene set compilation. Left panel: Arrary data processing. Affymetrix Exon 1.0 ST Array data was normalized, probe sets mapped to U133 plus 2 Array, and filtered based on redundancy, intensity, and coefficient of variation across all samples. Middle panel: IPF prognostic predictor gene set compilation. Three approaches used to compile IPF prognostic predictor gene set: Co-expressed gene modules correlated with pulmonary function identified by WGCNA; Differentially expressed genes between “good” and “poor” prognosis patients identified by SAM (fold change > 1.5 & FDR < 2.5 %); Survival-correlated genes identified by Cox regression (p < 0.005). Right panel: Genomic model IPF prognosis prediction. IPF prognostic predictor gene set was used to construct a genomic model; Prognostic Index (PI) score was calculated from each patient in training cohort; Prediction specificity was assessed by 10-fold cross validation; Genomic model was validated in two independent cohorts using weights of PI calculated from training cohort. b Venn diagram illustrates the selection criteria for IPF prognostic predictor genes. A total of 118 genes were compiled for downstream data analyses. c Canonical pathways enriched from IPF prognostic predictor genes by Ingenuity Pathway Analysis software. Significant pathways were set with criterion of q-value < 0.05 (i.e. -log (q-value) > 1.3) using one-tailed Fisher’s exact test. X-axis represents -log (q-value)