05). Functional gene enrichment analysis using DAVID resulted in 36 individual GO terms with significant enrichment. The non-redundant GO term set was subsequently visualized as tree map using REVIGO and the analysis revealed the superclusters “cell division” and “response to hormone stimulus” as major difference between the R2LC low and high risk groups ( Fig. 5). To assess expression levels of the selected biomarkers caveolin-1, NDKA, RPS6, and Ki-67 on the

transcript GSI-IX level, a comparison between mRNA and protein expression was carried out for 68 tumor samples of the discovery cohort limited to those tumors where mRNA data were available. Correlation analysis revealed that caveolin-1 mRNA and protein level were positively

correlated (p < 0.001) with a Spearman’s rank correlation coefficient of ρ = 0.646. NDKA and Ki-67 also showed a significant positive correlation (p < 0.001) with ρ = 0.682 and ρ = 0.402, respectively. In case of RPS6, no correlation between mRNA and protein expression was observed ( Fig. 6A). The recently published breast cancer data set of Curtis et al. [2] was used to compare gene expression levels of caveolin-1, NDKA, www.selleckchem.com/products/Gefitinib.html and Ki-67 with intrinsic molecular subtypes assigned to those samples using gene expression profiling data. In line with RPPA derived results, mRNA levels of caveolin-1 were significantly higher in luminal A compared with luminal B samples. In addition, NDKA and Ki-67 revealed a higher expression in luminal B samples (Fig. 6B). Breast cancer is nowadays recognized as a heterogeneous

disease with different intrinsic molecular subtypes. The luminal subgroup, which comprises the majority of cases, can be further divided into luminal A and luminal B associated with better or worse prognoses, respectively. This classification is crucial for therapy decisions as patients of the luminal B subtype with high risk of recurrence should be treated with chemo-endocrine oxyclozanide therapy whereas patients being at lower risk could be spared chemotherapy and its adverse side effects. However, a proper definition of low and high risk luminal breast cancer to aid treatment decisions has so far remained a challenge. This study identified a protein biomarker signature consisting of caveolin-1, NDKA, RPS6, and Ki-67 by using RPPA-based tumor profiling which should improve determining the recurrence risk in patients with luminal breast cancer. Biomarker selection was based on a new bioinformatics approach, bootfs, firstly introduced here. Bioinformatics offers numerous methods to solve two-group classification problems in high-throughput data sets. However, no approach clearly outperforms any other algorithm for all quality criteria at once, namely prediction accuracy, feature selection stability, and biological relevance [ [31] and [32]].