Transcriptomic analysis using SVD clustering and SVM classification

Abstract

The classification performance using support vector machines (SVMs) for transcriptomic analysis can be limited due to the high dimensionality of the data. This limitation is most problematic in the case of small training sets. A general solution is to employ a dimension reduction method before SVM classification. In this paper, we propose a novel singular value decomposition (SVD) based method for dual purposes: firstly, to reduce the dimensionality, and secondly to cluster the transcriptional profiles. The kernel functions of SVM were modified based on the Riemannian geometrical structure which can achieve a better spatial resolution. The proposed approach was applied to the yeast time series microarray dataset and outperformed the traditional SVM kernels.