Identifying biological mechanisms for favorable cancer prognosis using non-hypothesis-driven iterative survival analysis.

Abstract

Survival analyses based on the Kaplan–Meier estimate have been pervasively used to support or validate the relevance of biological mechanisms in cancer research. Recently, with the appearance of gene expression high-throughput technologies, this kind of analysis has been applied to tumor transcriptomics data. In a ‘bottom–up’ approach, gene-expression profiles that are associated with a deregulated pathway hypothetically involved in cancer progression are first identified and then subsequently correlated with a survival effect, which statistically supports or requires the rejection of such a hypothesis. In this work, we propose a ‘top–down’ approach, in which the clinical outcome (survival) is the starting point that guides the identification of deregulated biological mechanisms in cancer by a non-hypothesis-driven iterative survival analysis. We show that the application of our novel method to a population of ~2,000 breast cancer patients of the METABRIC consortium allows the identification of several well-known cancer mechanisms, such as ERBB4, HNF3A and TGFB pathways, and the investigation of their paradoxical dual effect. In addition, several novel biological mechanisms are proposed as potentially involved in cancer progression. The proposed exploratory methodology can be considered both alternative and complementary to classical 'bottom–up' approaches for validation of biological hypotheses. We propose that our method may be used to better characterize cancer, and may therefore impact the future design of therapies that are truly molecularly tailored to individual patients. The method, named SURCOMED, was implemented as a web-based tool, which is publicly available at http://surcomed.vital-it.ch. R scripts are also available at http://surcomed.sourceforge.net).