Development and integrative analysis of a cancer gene expression database.

Abstract

DNA microarrays have been widely used to study gene expression in human cancer. The primary analyses to date highlight the molecular heterogeneity of cancer. Integrative analyses linking data on gene expression patterns in cancer with other biological and clinical data are expected to further knowledge of cancer pathogenesis and aid in defining novel biomarkers. To integrate and mine the published compendium of cancer gene expression data, we developed Oncomine (www.oncomine.org), a database and web application encompassing more than 15,000 tumor profiles. To obtain a systems perspective of cancer biology from Oncomine, integrative computational analysis strategies were developed. A first methodology, meta-analysis of microarrays, was used to compare independent microarray datasets generated by distinct microarray platforms. This approach facilitated integration of four prostate datasets and the identification of multi-study validated prostate cancer biomarkers. The methodology was also applied to 40 datasets representing 12 cancer types, allowing identification of a gene expression program activated in essentially all cancer types. A second strategy involved predicting protein-protein interactions from diverse genomic and proteomic data sources to generate an in-silico human protein interaction network, facilitating identification of protein networks deregulated in cancer. A third methodology integrated transcription factor binding site data with cancer gene expression signatures to predict transcriptional programs activated in cancer. Notably, widespread activation of the E2F program in several aggressive cancer types was identified. A fourth analysis methodology, inspired by the heterogeneity of cancer, was used to scan Oncomine for candidate oncogene expression profiles. This approach identified known oncogenes deregulated in cancer and nominated candidate oncogenes, including ERG and ETV1 in prostate cancer, which were both found to be activated by gene fusion events. Lastly, to integrate and link diverse genomic signatures, the Molecular Concept Map was developed. This methodology utilizes complementary data sources and enrichment analysis to identify the processes, pathways and networks deregulated in cancer. In summary, the development of Oncomine combined with biology-driven computational analysis methodologies has yielded new insights into cancer pathogenesis with important ramifications for uncovering new biomarkers and therapeutic targets.