Sample-level enrichment analysis unravels shared stress phenotypes among multiple cancer types

We propose a new approach based on enrichment analysis at the level of samples (sample-level enrichment analysis - SLEA) in expression profiling datasets. Without using a priori phenotypic information about samples, SLEA calculates an enrichment score per sample per gene set using z-test. This score is used to determine the relative importance of the corresponding pathway or module in different patient groups.Our analysis shows that tumors significantly upregulating genes related to chromosome instability strongly correlate with worse prognosis in breast cancer. Moreover, in multiple tumor types, these tumors upregulate a senescence-bypass transcriptional program and exhibit similar stress phenotypes.Using SLEA we are able to find relationships between stress phenotype pathways across multiple cancer types. Moreover we show that SLEA enables the identification of gene sets in correlation with clinical characteristics such as survival, as well as the identification of biological pathways/processes that underlie the pathology of different cancer subgroups.Complex genetic diseases such as cancer are characterized by phenotypic heterogeneity reflected at the molecular level in the form of variations in the activity of certain signaling pathways. In support of this notion, recent cancer genome studies point to the idea that distinct types of alterations in different genes tend to accumulate in pathways central to the control of cell growth and cell fate determination [1-4]. It has been proposed that expression signatures indicative of activity status of pathways can be used to define specific molecular phenotypes that characterize individual tumors [5]. A number of methods have been developed to analyze the transcriptomic changes specific to tumor samples and identify patterns of pathway deregulation that differentiate distinct patient subgroups [6-12]. These methodologies are based on the idea that analysis of pathway-level differences among samples could have an advant