Ovarian Cancer Identification Based on Feature Weighting for High-Throughput Mass Spectrometry Data
An important use of proteomics data from Mass Spectrometry (MS) is the classification of tumor types with respect to peptides in specific cancer types. It is highly critical to find an optimal set of markers among specific cancer peptides whose expression can be clinically utilized to build assays for the diagnosis or to track the progression of specific cancer types. A number of feature selection algorithms have been proposed to obtain the classification of MS data. In this article, we proposed an improved feature selection algorithm based on feature weighting. Relief algorithm can calculate the weight of different features according to the correlation between their characteristics and categories. F-score is a simple filter-based feature selection method by evaluating how two sets of real numbers discriminate from each other. The main goal of this paper is to introduce a new feature weighting selection algorithm combining score from f-value and weight from relief, which is more accurate when classifying high-resolution MALDI-TOF (matrix-assisted laser desorption and ionization time-of-flight) MS data. We have developed a four-step strategy for data processing based on: (1) Align the study sets by binning of raw MS data, (2) local maximum search(LMS) peak detection, (3) a new combination feature weighting selection algorithm and (4) support vector machines achieve a satisfactory performance of identifying cancer and the healthy. The best parameter set for LMS were achieved with control variable method, which achieve an average accuracy of 97.4167% (sd = 0.0146) and the best accuracy of 98.6111% in 1000 independent 10 -fold cross validations.