Network-based identification of reliable bio-markers for cancers

Finding bio-markers for complex disease from gene expression profiles attracts extensive attentions for its potential use in diagnosis, therapy, and drug design. In this paper we propose a network-based method to seek high-confident bio-markers from candidate genes collected in the literature. The algorithm includes three consequent steps. First, one can collect the proposed bio-markers in literature as being the preliminary candidate; Second, a spanning-tree based threshold can be used to reconstruct gene networks for normal and cancer samples; Third, by jointly using of degree changes and distribution of the candidates in communities, one can filter out the low-confident genes. The survival candidates are high-confident genes. Specially, we consider expression profiles for carcinoma of colon. A total of 34 preliminary bio-markers collected from literature are evaluated and a set of 16 genes are proposed as high confident bio-markers, which behave high performance in distinguishing normal and cancer samples.