Normalization of gene expression data using support vector machine approach

Abstract

Normalization of gene expression data refers the process of minimizingnon-biological variation in measured probe intensity levels so that biologicaldifferences in gene expression can be appropriately detected. Several linearnormalization within arrays approaches have already been proposed. Re-cently, use of non-linear methods has been gained quite attention. In thisstudy, our objective is to formulate non-linear normalization methods usingsupport vector regression (SVR) and support vector machine quantile re-gression (SVMQR) approaches more easier way and, assess the consistencyof these methods with respect to other standard ones for further applicationin gene expression data. After implementation, the performances of SVRand SVMQR have been compared with respect to other standard normal-ization methods namely, locally weighted scatter plot smoothing and kernelregression. The results indicate that the normalized data based on proposedmethods are capable of producing minimum variances within replicate groupsand, also able to detect truly expressible significant genes compared to abovementioned other normalized data.