A trend pattern assessment approach to microarray gene expression profiling data analysis

We study the problem of how to assess the reliability of a statistical measurement on data set containing unknown quantity of noises, inconsistencies, and outliers. A practical approach that analyzes the dynamical patterns (trends) of the statistical measurements through a sequential extreme-boundary-points (EBP) weed-out process is explored. We categorize the weed-out trend patterns (WOTP) and examine their relation to the reliability of the measurement. The approach is applied to the processes of extracting genes that are predictive to BCL2 translocations and to clinical survival outcomes of diffuse large B-cell lymphoma (DLBCL) from DNA Microarray gene expression profiling data sets. Fisher's Discriminate Criterion (FDC) is used as a statistical measurement in the processes. It is found that the weed-out trend analysis (WOTA) approach is effective for qualitatively assessing the statistics-based measurements in the experimentations conducted.