Detection Of Differentially Expressed Genes

A central aim in cancer research is to identify altered genes (over- or down- expressed genes ) that play a causal role in cancer development.mRNA microarray experiments provide a fast and systematic way to identify disease markers relevant to clinical care. A common experimental design is the comparison of two sets of samples from different phenotypes (diseases and normal tissue), with the goal of searching for genes showing differential expression. This is usually done via statistical testing procedures and, often, subsequent multiple testing corrections. All these methods use a one-gene-at-a-time strategy, considering only the association between single genes and the phenotype.The identification of differentially expressed genes is a question which arises in a broad range of microarray experiments. Many methods have been proposed to detect differentially expressed genes Dudoit and others [5]. Among them, t-statistics is the most commonly used method.In the majority of cancer types, heterogeneous patterns of oncogene activation have been observed; thus, traditional analytical methods that search for common activation of genes across a class of cancer samples (e.g., t test) will fail to find such oncogene expression profiles. Instead, a method that searches for marked over-expression in a subset of cases is needed.Recently, Tomlins and others [4] have proposed the "cancer outlier profile analysis" (COPA) method for detecting cancer genes which show increased expressions in a subset of disease samples. More recently, Tibshirani and Hastie [1] proposed the outlier sum (OS) statistic to detect cancer gene outlier expressions. The OS and COPA are similarly defined using robust location and scale estimates of the gene expression values. Later Wu [2] proposed outlier robust t-statistics (ORT) and LIAN [3] propose another statistics MOST: maximum ordered subset t-statistics which, through simulation studies shows that the MOST can perform generally better than all the previous statistics.We will discuss the performances of the above statistic for cancer gene outlier expression detection, through some simulation process, which will motivate the development of our improved methods. And we will further show the connection of the MOST, ORT, OS, COPA, and the t-statistic to the improved methods from a robustness consideration view.