Minimum Redundancy Maximum Relevance for Analysis of Proteomic Profile

Abstract

We propose a approach to determine features directly based on classification target of proteomic profile, which combines minimum redundancy maximum relevance (MRMR) and support vector machines (SVM). Firstly, the profile are preprocessed through iterative minimum in adaptive setting window (IMASW) and searching window methods for correcting negative intensities caused by manual preprocessing and peak picking. Then, MRMR and support vector machines (SVM) techniques were used to identify biomarkers and build discrimination model. With an optimization of the parameters involved in the modeling, a satisfactory model was achieved for ovarian cancer diagnosis based on proteomic peptide profile dataset. To study the performance of MRMR, we applied two statistical method, t test and Wilcoxon signed-rank test, to identify features. The results show that MRMR method is more efficient.