Extracting key pathways from gene signature and genetic aberrations in subtypes of cancer
Subtypes of cancer are characterized with subtype-specific aberrations and gene signature. While the gene signature is related to the consequences of the cancerous process, some of the genetic abnormalities such as copy number aberrations (CNAs) can have tumorigenic roles by perturbing various biological pathways. Bridging the gap between the aberrations and signature genes, by extracting networks that reflect the within-subtype variations, may help gain insights on the mechanisms of a cancer and its subtypes. We report a systemic approach to extract pathways. Using multivariate regression, we model the expression of a signature gene as dependent on the CNA-affected genes. The weighted ℓ1-norm penalty on the regression produces a sparse matrix, from which a bipartite graph is extracted and subtype specific networks uncovered. For each individual network, we develop an network-growing algorithm by utilizing within-subtype variations, to further identify non-signature targets. To evaluate the clinical relevance of the extracted networks, we derived a goodness-of-fit metric based on Cox proportional hazard rate model and ranked the networks based on this metric. The method was applied to two medulloblastoma datasets and the resulting networks demonstrate both dataset-invariance and biological-interpretability. © 2013 Springer-Verlag.
Digital Object Identifier (DOI)
Additional Document Info