Abstract 52: A new method to interpret gene expression microarray data for individual tumors by profile alignment against a reference dataset of 10,000 samples

Abstract

Abstract Transforming genome wide expression profiles into meaningful biological insights is a complex problem, particularly for individual tumor samples. Numerous data-analysis methods have been developed for this purpose, but the vast majority of these aim at the identification of large subgroups of tumors based on patterns of similarity in gene expression within the investigator's own dataset. There are now several large microarray databases (e.g. GEO, Array-express, Oncomine, GeneSapiens) and meta-analysis projects that could be helpful in the interpretation of single microarray data. Our aim was to develop a microarray profile alignment method that would serve the exploration of transcriptomics data in analogy to how “BLAST” analysis is routinely performed for nucleotide sequence comparisons. Here, we describe a tool to facilitate the interpretation and exploration of microarray data from individual tumors with the help of the large uniform reference database, Gensapiens (www.genesapiens.org). The tool enables us to compare a single sample with any tissue or tumor type in the database, and determine the similarity between them. A leave-one-out cross validation test with the Genesapiens database samples achieved 93.6% classification accuracy for 47 normal tissue types and 87.5% for cancer tissue of origin, among 57 different cancer types. The latter result suggests application of the method in the diagnosis of unknown primary tumors (CUPS). We applied the method to a set of metastasis samples and achieved an overall accuracy of 75%, which could further be refined to 88% by discarding uncertain results, that is, leaving out the samples that have no significant similarity to any cancer. The similarity of the test sample to a tissue or cancer type can be further explored at the level of individual genes, gene sets, gene ontology subsets or pathways. As an example of personalized medicine applications, we aligned datasets from two external metastatic (liver and lung) samples and correctly identified both cases as colorectal cancer. However, 3-4 % of the genes had expression levels rarely seen in colorectal cancer, highlighting unique features and potential individual biomarkers to these two tumors. In conclusion, we have developed a gene expression microarray profile alignment tool that provides a new opportunity for the interpretation of microarray data from individual tumor samples against a large reference database, such as GeneSapiens. This could be helpful in i) independent validation of the tumor type ii) subclassification of the sample in relation to a large reference dataset, iii) exploring unique features of the tumor and iv) guiding personalized biomarker discovery and therapy. Thus, this provides a new paradigm for the analysis of microarray data and an ability to uncover biological insights from individual gene expression profiles. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 52. doi:10.1158/1538-7445.AM2011-52