Identification of a tumour percentage transcriptional profile. (A) Classifier performance (area under ROC curve) using different top-ranked gene set sizes. Performance is given for classification of low (<30%) and high (≥50%) TCP samples. (B) Differential gene expression of top 35 ranked genes between low and high TCP samples. Bars represent expression variation of each gene across 70 high TCP samples. Thirteen most stably differential expressed genes for high TCP are indicated in blue. (C) Heat-map of the 13-gene tumour percentage profile for 95 training samples. Samples are ordered according to their pathologist scored tumour percentage. (D) Profile outcome for all 165 training samples. Profile indexes are calculated using a leave-one-out CV. Tumour samples are grouped according to their tumour percent scoring and ordered according to the profile index. Horizontal dashed line represents the optimal classification threshold.