A Robust and Efficient Statistic for Detecting Heterogeneous Cancer Samples

• Activity Number: 634
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Katherine Thompson*+ and Shili Lin
• Companies: The Ohio State University and The Ohio State University
• Keywords: differential gene expression ; heterogeneous cancer samples ; efficient statistics ; robust statistics

Abstract

It is known that particular genes influence cancer development through either over or under expression. More recently, research has shown that some genes are differentially expressed in only a portion of cancer samples. Being able to detect this differential expression could prove useful in the diagnosis and treatment of cancers. Using the traditional t-test tends not to be robust when there is heterogeneity. On other other hand, recent methods (OS, ORT, and COPA statistics) are robust but not as efficient. Our statistic is designed so that it achieves the efficiency of the t statistic (when disease samples contain homogeneously over or under expressed genes), while it preserves the robustness of other statistics (when disease samples are heterogeneous). We compare the performance of our statistic to that of the other methods discussed above. We then propose a method using this statistic to perform a hypothesis test for differential expression heterogeneity in cancer samples.