Identification of Breast Cancer Gene Signature in Protein Interaction Network Using Graph Centrality

Various gene-expression signatures for breast cancer are available for prediction of clinical outcome, but due to small overlap between different signatures, it is challenging to integrate existing disjoint signatures to provide a unified insight on the association between gene expression and clinical outcome. In this paper, we proposed a method to identify reliable breast cancer gene signature from a context-constrained protein interaction network(PIN). The context-constrained PIN for breast cancer is built by integrating complete PIN and various gene signatures reported in literature. Then, we used graph centrality to quantify the importance of genes to breast cancer. Finally, we got reliable gene signatures that are consisted by the genes with high graph centrality. The genes which are well- known breast cancer genes, such as TP53 and BRCA1 are ranked extremely high in our results. Compared with previous result by functional enrichment analysis, graph centrality, especially the eigenvector centrality and subgraph centrality based gene signatures are more tightly related to breast cancer. We validated these signatures on genome-wide microarray dataset and found higher relationship between the expression of these signature genes and pathologic parameters. In summary, graph centrality provides a novel way to connect different cancer signatures and to understand the mechanism of relationship between gene expression and clinical outcome of breast cancer. Moreover, this method is applied not only to breast cancer, but also to other gene expression related diseases.