Statistical discrimination of breast cancer microarray data

Abstract

Recently published literatures indicate that, a consistent and precise classification system of microarray data is very essential for the successful identification of genes responsible for a cancer subtype and early treatment of cancer. However, in common clinical practice, diagnostic assertion of malignancies mostly relies on the morphological examination of tissues and clinical tests. In spite of the recent progress of treatment that uses semiempirical approaches involving computation methods, there are still uncertainties in cancer identification and diagnosis. High density oligonucleotide chips and genomic microarray data are being used progressively more in cancer research by generating huge expression data for thousands of genes simultaneously that poses problem of mining specific genes responsible for characterization of a specific cancer subtype. In this backdrop, we briefly address the impact of various statistical methods and their relative performances for the identification of potential probes and Discriminant genes for the breast cancer. We used Fisher's Discriminant Ratio (FDR), two tailed T-Test and vector norm on raw expression data and for each of the probe generated from the difference data accounting for up and down regulation of expression for each probes for various samples. The result indicates the potential of these methods to identify genes responsible for manifestation of breast cancer which is also well supported by the published result of experiments. The success of this approach not only gives the benefit of identification of cancer specific genes but also may help building of efficient classifier on the basis of these genes for automatic diagnostics of cancer.