Prioritizing Context-Dependent Cancer Gene Signatures in Networks

There are numerous ways of portraying cancer complexity based on combining multiple types of data. A common approach involves developing signatures from gene expression profiles to highlight a few key reproducible features that provide insight into cancer risk, progression, or recurrence. Normally, a selection of such features is made through relevance or significance, given a reference context. In the case of highly metastatic cancers, numerous gene signatures have been published with varying levels of validation. Then, integrating the signatures could potentially lead to a more comprehensive view of the connection between cancer and its phenotypes by covering annotations not fully explored in individual studies. This broader understanding of disease phenotypes would improve the predictive accuracy of statistical models used to identify meaningful associations. We present an example of this approach by reconciling a great number of published signatures into meta-signatures relevant to Osteosarcoma (OS) metastasis. We generate a well-annotated and interpretable interactome network from integrated OS gene expression signatures and identify key nodes that regulate essential aspects of metastasis. While the connected signatures link diverse prognostic measurements for OS, the proposed approach is applicable to any type of cancer.