Björn Samuelsson Protein Maps Based on Structural Alignments Master Thesis Advisor: Markus Ringnér Abstract: A method to map protein domains to a Euclidean space, using the RMS-distances given by pairwise structural alignments, is presented. First, the domains are mapped onto points in a space with a generalised scalar product such that all distances are conserved. Next, the points are restricted to the largest Euclidean subspace, by elimination of dimensions in a smooth way. Further dimensions are eliminated until the space has as few dimensions as possible without getting too large deviations from the original RMS-distances. The method is tested for a set of globin domains and a set of all-alpha domains. For the globins, the procedure gives a clear and biologically relevant grouping, but for the set of all-alpha domains the structure is much fuzzier. Some simple clustering algorithms are also tested on the globin domains. October 2000 LU TP 00-43