Abstract 42: Potential colon cancer biomarkers: Multiple criteria gene selection with microarray data

Abstract

Abstract In this work, the simultaneous analysis of two publicly-available affymetrix microarray databases for colon cancer is carried out with the objective to detect potential biomarkers. Gene selection is approached here through its formulation as a multiple criteria optimization (MCO) problem. This formulation allows (i) removing the need for parameter adjustment by the user -thereby preserving the objectivity of the results in terms of gene selection-, and (ii) analyzing several databases with different characteristics in a concurrent manner. The key idea to adopt an MCO approach is the representation of each gene under analysis through multiple performance measurement values. The performance measures in this work are statistical p_values obtained through nonparametric statistical comparison. A gene with minimal p_values across all instances can be deemed to be a potentially dominant gene. However, due to the nature of the analysis based on experimental data, the simultaneous consideration of multiple p_values will lead to conflicting behavior among them. With this in mind, the objective of the MCO problem is to identify the genes that are demonstrably the best compromises among all genes. These best compromises are termed efficient solutions and are located in the so-called Pareto efficient frontier of the optimization problem. An efficient solution can improve in a performance measure only when worsening at least another performance measure. The concurrent analysis of the afore-mentioned microarray databases results in the consistent identification of 13 genes, all of which were found to have roles on cancer in our literature search and review. Among the thirteen genes detected through the application of our method, the tumor suppressor activator GSN and the metastasis suppressor NME1 showed significantly different expression levels in cancer tissues when contrasted to normal tissues. HSPD1, which has been already reported as potential biomarker gene for prostate cancer, was also included in the final selection. GTF3A, DES, NPM1, CSRP1 and MT1G were picked by the analysis and have been reported to significantly change their expression levels in other cancers, albeit not specifically in colon cancer. MYL9 and VIP have not been reported as potential cancer biomarkers, however, their cellular roles indicate that they might be related to the illness. Finally, completing the list, CFD, GUCA2B and one EST have not been formally related to cancer. These last five genes could represent important research opportunities in the search for colon cancer biomarkers. These results allow appreciating two important qualities of the proposed method: (i) high discrimination rate and (ii) independence of parameter adjustment by the user. Its use for secondary analyses of existing microarray databases is proposed to enhance the objectivity of the identification of potential biomarkers. 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 42. doi:10.1158/1538-7445.AM2011-42