Partner und Internationale Organisationen
(Englisch)
|
AT, BE, BG, CY, EE, FR, DE, IE, IT, MT, NO, PL, PT, SK, ES, CH, UK
|
Abstract
(Englisch)
|
A major research topic in functional genomics and proteomics is the search for biomarkers, or meaningful indicators of biological states or processes. Of special interest are disease biomarkers, which can be used for disease diagnosis, monitoring, and control. Genomics-based approaches have made it possible to identify changes in gene expression levels as markers of certain diseases. More recently, there has been increasing interest in the proteome which, contrary to the genome, reflects both the organism's genetic blueprint and the impact of its environment. As a consequence, mass spectrometry is emerging as a vital tool for biomarker discovery. Mass spectra of biological samples such as serum are analysed using advanced data mining techniques in view of extracting diagnostic protein patterns. To extract biomarker patterns from protein mass spectra, data miners must face a technological challenge which has become the focus of research in the FunProt project. This is the so-called high-dimensionality-for-small-sample (HDSS) or the p >> n problem -- that is, the number of variables (p) is much larger than the number of cases (n). Most data analysis algorithms fail to generalize or break down completely in the presence of datasets where p >> n. To tackle this challenge, two alternative approaches have been explored in FunProt. The first is preliminary dimensionality reduction: prior to data mining, the number of variables or features is reduced via feature (subset) selection, feature clustering, or feature transformation. The second is the investigation of learning algorithms that are either inherently insensitive to the HDSS problem (e.g. support vector machines), or that comprise built-in mechanisms for reducing the number of variables on the fly (e.g. shrunken centroids). Project work has led to the extraction of biomarker patterns for Alzheimer's disease and cerebrovascular accidents in collaboration with the Biomedical Proteomics Research Group of the Faculty of Medicine. Research results have been presented in artificial intelligence conferences and in major specialized journals such as Proteomics, Biosilico, and Mass Spectrometry Reviews.
|
