Figure 1: Hierarchical clustering of infected and control patient samples.

(a) Results from unsupervised analysis of normalized data before statistical testing. At this stage, the condition of whether the samples were infected or not had not been included in the analysis. Control, blue; infected, red. (b) Heat map showing hierarchical clustering of infant samples based on the 824 probes that were statistically differentially expressed between infected and control groups (using eBayes with Benjamini-Hochberg correction adj.P≤0.01, absolute fold change ≥2). Hierarchical clustering was based on Euclidean distance. Three clusters of genes are labelled (1–3). Group 3 genes (highlighted with an orange box) were excluded from the further analysis for derivation of a classifier of neonatal bacterial infection because of no clear separation between infected and controls. Control, blue; infected, red.