Cancer classification from oligonucleotide arrays using gene dependent estimators

Abstract

Conventional histopathological examinations have been known to be unreliable in cancer diagnosis. Upon diagnosing cancer, specific clinical treatments have to be sought based on the class of cancer detected. Therefore, cancer classification is imperative. The DNA microarray technology is a promising technology that could revolutionize the way cancers are diagnosed. State-of-the-art cancer classification techniques suffer from the problems posed by the presence of outliers, irrelevant genes, and inter-dependent genes. This dissertation puts forward a machine learning based framework that could recognize and classify cancer from oligonucleotide gene expression data. To circumvent the problems faced by the state-of-the-art cancer classification systems, this dissertation employs an entropy-based transcriptomic marker selection approach to select oncogenes and relevant marker genes that are directly responsible for cancer discrimination. An entropic transcriptome discretization technique is utilized in order to alleviate the effect of outliers and increase the generalization capability of the system. The proposed system was found to outperform the state-of-the-art systems by circumventing the fundamental problems caused by the state-of-the-art systems. The results demonstrate the efficacy of the proposed cancer classification framework. The proposed framework can be applied in various areas of clinical oncology.