Extended Study in Computing Science 20p Pattern Recognition Algorithms for Cancer Detection applied on data from SELDI Technology

The SELDI technology is a technique that measures the content of different proteins in blood samples from patients. Many research groups have shown that there appears to exist a relation between the concentrations of speci c proteins and cancer disease. The output from the SELDI system is a result of mass-spectrometry and is a spectrum containing the concentrations of thousands of separate proteins. This report is an extended study based on the master thesis project "Analysis of Proteomic Patterns for Detection of Prostate Cancer" [17], in which di erent methods for classi cation of SELDI spectra were evaluated. The most promising investigated algorithm for this task was Fishers Linear Discriminant (FLD). During the present project, the pattern recognition methods Support Vector Machines (SVM) and Multilinear Regression (MLR) have been studied for the same purpose and have been compared to FLD with respect to classi cation performance. The conclusions of this study are that MLR and SVM both performs as well as FLD for the available data sets and that MLR is the fastest algorithm with respect to computational complexity. Before these algorithms can be e ciently applied to the SELDI spectra, the dimensionality of the samples must be reduced. The most e ective known way to do this is to use Principal Component Analysis (PCA). This study describes PCA and also introduces Random Projection (RP) as an alternative method. The report also discusses some unwanted features of the SELDI system which were discovered during an extensive prostate cancer study performed by Pernilla Wikström and Åsa Skytt at the Department of Bioscience, Pathology, Umeå University. These features have been known the recent time and have caused some previously published results to be questioned by di erent research groups working in the area of proteomics. Umeå University Department of Computing Science 1 Pattern Recognition Algorithms for Cancer Detection applied on data from SELDI Technology Umeå University Department of Computing Science 2 Pattern Recognition Algorithms for Cancer Detection applied on data from SELDI Technology