Genetic algorithm based feature selection for mass spectrometry data

Mass spectrometry technique is a revolutionary tool for diagnosing early stage cancer by analyzing protein mass spectra, and for detecting biomarkers. But because of the high dimensionality of the data, feature selection is a necessary procedure before classification and analysis. In this paper we present a genetic algorithm for feature selection for prostate protein mass spectrometry data. An elitism coupled with rank based stochastic universal sampling selection strategy, uniform crossover operation, and a uniform mutation with adaptive mutation rate strategy are used. Two fitness functions are defined for the genetic algorithm: one is a multivariate filter measurement and the other is a wrapper measurement. Our experiments show that the wrapper-based genetic algorithm outperforms all the other feature selection methods presented here. The multivariate filter-based genetic algorithm also yields better performance than transformed methods, sequential selection methods, and univariate filter methods.