Combining singular value decomposition and t-test into hybrid approach for significant gene extraction from microarray data

Significant gene extraction from microarray data is a challenging problem which is of great interest to researchers in Computational Biology, Medicine, Computer Science and Statistics. A number of methods have been proposed for extracting the smallest number of genes which can accurately classify different samples. Most of these methods ignore the fact that microarray data is mostly noisy. For instance, only using a statistical t-test has been shown to be insufficient since it result in a high false discovery rate. Recently, a singular value decomposition (SVD) based approach was proposed for time series microarray data reduction, however it turned out not to be efficient for classifying microarray data. To overcome the shortcomings of these approaches, this paper proposes two methods to reduce false discovery rates. The first method involves an iterative t-test which finds the p-value for each gene under perturbation by eliminating one sample at a time. It eliminates weak noisy genes by dropping any gene which does not show significant p-value under all the conditions. The second method is a hybrid process which adapts a combination of the SVD and the t-test. It considers the entropy of all the data, and thus takes the correlation between genes into account. Classification accuracy is used to validate the significance of the extracted genes. The reported test results on two datasets demonstrate the applicability and effectiveness of the two proposed methods.