Improvement of SVM Algorithm for Microarray Analysis Using Intelligent Parameter Selection

Identification of genetic markers is a crucial step in the diagnosis, prognosis, and treatment of disease. This paper focuses on the application of a supervised classification technique, support vector machines (SVM), to high dimensional microarrays for marker identification. A case study of renal cell carcinoma (RCC) is used here to demonstrate and test the ability of SVMs to identify real biological markers. SVMs are known to be suitable for high dimensional microarray data and are able to classify non-linear relationships in the data through the use of kernel functions specific to the datasets. This paper compares multiple SVM kernel functions, both linear and nonlinear, to determine which form is best suited for a particular dataset. Additionally, each SVM is tested across a range of parameters and normalization schemes to further identify a specific optimal classifier. Empirical results are then used to determine the optimum parameters for the SVM to efficiently find biologically important predictive markers for differentiation between RCC subtypes for the purpose of diagnosis and prognosis