Extraction and selection of relevant information in proteomic mass spectra for biomarker identification.

摘要

A diagnostic decision of a physician in often based on measurements of some quantities in body tissue or fluids. Those measurements serve as markers of specific diseases and their stages. Biomedical research is constantly seeking for markers to diagnose diseases earlier and with better precision. Proteins have been recently regarded as very promising markers of cancer and many other diseases. Mass spectrometry combined with pattern recognition and data mining methods has become a very popular tool in discovery of proteomic biomarkers. Due to high noise levels, large dimensionality and small number of available examples, automatic analysis of proteomic mass spectra is still a very challenging task. The spectra were in many cases successfully classified into diseased and control groups with promising generalization estimates. However the identified differences between the studied groups of the spectra did not allow for identification of proteins that exist only in one group. Thus automatic analysis and discovery of biomarkers in proteomic mass spectra is still an open research topic. The work presented in this dissertation analyzes techniques developed in the areas of pattern recognition, machine learning, and data mining with respect to applicability to processing of proteomic mass spectra. The goal of this dissertation is to propose techniques that allow for more efficient and automatic search for biomarkers in proteomic mass spectra.