Abstract 22: Identifying chemotherapy resistance genes using outlier detection

Abstract

Abstract Motivation: Breast cancer is a heterogeneous disease and although gene expression analysis or immuno histo chemistry can be used to subclassify tumors into distinct subgroups, even within these subgroups there exist significant differences at the molecular level. When it comes to chemotherapy resistance, it appears highly unlikely that all resistant tumors share the same mechanism of action. Classical approaches have been unsuccessful in finding reliable biomarkers for therapy resistance and this may be due to the fact that these approaches are not sensitive to changes in small subgroups of samples. Methods: Traditional analyses approaches, like a t-test or using the SAM approach, compare the two groups of samples for each gene, and identify genes with a significant difference in expression between the two groups. If only a small subset of the resistant tumors would show aberrant expression indicative of a resistance mechanism, then this would not be picked up by such approaches. Using a novel algorithm we try to circumvent this problem and specifically aim to find relatively small subgroups of tumors within the resistant group that show differential expression compared to the sensitive group. Results: Using a positive control set, where we mixed in a small set of HER2-positive tumors in a larger HER2-negative set and by generating an artificial dataset we show that our algorithm is able to pick up all positive controls. These controls are not picked up by either a t-test or the SAM algorithm. Next, we applied the algorithm on a gene expression set of 195 patients that were neoadjuvantly treated and identified a number of genes that show an expression pattern that suggests a role in chemotherapy resistance in ER+ tumors. We validate our findings on a separate set of 90 patients. Discussion: We have developed a novel algorithm that allows the identification of genes that show aberrant expression, relative to the reference group, in a small subset of the samples. Using this approach we identified a number of genes that are linked to chemotherapy resistance. The algorithm can be used in any type of analysis that involves the detection of small subsets of samples within one of the labeled classes, where the small subset shows behavior different from the average behavior of the remaining samples. Our approach can therefore be applied to a wide range of problems. 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 22. doi:10.1158/1538-7445.AM2011-22