Classification Using Mass Spectrometry Proteomic Data with Kernel-Based Algorithms

Motivation: Early detection of cancer is crucial for successful treatment, and protein profiling using mass spectrometry (MS) data has been investigated as a potential tool. However, due to the high correlation and huge dimensionality of MS data, it is crucial to modify existing algorithms and to develop new ones, where necessary, for analyzing such data. Results: We develop a group of logistic regression kernel coupling algorithms for classification of normal versus cancer samples. Furthermore, we propose a systematic three-step protocol for analyzing MS data, from removal of baseline noise and normalization, to feature extraction and reduction, and finally to classification. The systematic analysis paradigm and the proposed algorithms were applied to an ovarian and a prostate cancer dataset. The results show that our proposed approaches can be an effective tool for analyzing MS proteomics data.