Feature selection and machine learning with mass spectrometry data.

Mass spectrometry has been used in biochemical research for a long time. However, its potential of discovering proteomic biomarkers using protein mass spectra aroused tremendous interest in last few years. In spite of its potential of biomarker discovery, it is recognized that identification of meaningful proteomic features from mass spectra needs careful evaluation. Hence, extracting meaningful feature(s) and discriminating the samples based on these features is still an open area of research. Several research groups are actively involved in making the process as perfect as possible. In this chapter, we provide a review of major contributions toward feature selection and classification of proteomic mass spectra involving MALDI-TOF and SELDI-TOF technology. Moreover, in this updated version of the chapter, we advocate the use of an adaptive ensemble classifier to classify such complex data. No single classification algorithm tends to work well on all data. Also, the performance depends on the performance criteria used to judge several classifiers. Adaptive ensemble classifier which is constructed combining several good classifiers and optimized against an array of performance measures tends to have better predictive performance on the test samples.