Abstract—Recent advances in microarray technology offer the ability to study the expression of thousands of genes simultaneously. The DNA data stored on these microarray chips can provide crucial information for early clinical cancer diagnosis. The Principal Orthogonal Decomposition (POD) method has been widely used as an effective feature detection method. In this paper, we present an enhancement to the standard approach of using the POD technique as a disease detection tool. In the standard method, cancer diagnosis of an arbitrary sample is based on its correlation value with the cancerous or normal signature extracted using the POD method from DNA microarray data. In this paper, we extend the POD method by feeding the extracted principal features into Machine Learning algorithms to detect cancer. Particularly, Linear Support Vector Machine, Feed Forward Back Propagation Networks, and Self-Organizing Maps are used on liver cancer, colon cancer, and leukemia data. Sensitivity, specificity, and accuracy are discussed to evaluate predictive abilities of the proposed extended POD methods. Our results indicate overall the proposed methods provide improvements over the standard POD method.

Index Terms—DNA Microarray, principal orthogonal decomposition, machine learning, artificial neural networks, support vector machine, self-organizing map, cancer detection

Carlyn-Ann B. Lee is with the Department of Computer Science at California State University, Fullerton, CA 92834, USA (e-mail: cblee@ csu.fullerton.edu).
Charles H. Lee is with the Department of Mathematics at California State University, Fullerton, CA 92834, USA (e-mail: CharlesHLee@fullerton.edu).

Cite: Carlyn-Ann B. Lee and Charles H. Lee, "Extended Principal Orthogonal Decomposition Method for Cancer Screening," International Journal of Bioscience, Biochemistry and Bioinformatics vol. 2, no. 2, pp. 136-141, 2012.