Fober, ThomasMernberger, MarcoMoritz, RalphHüllermeier, EykeGrosse, IvoNeumann, SteffenPosch, StefanSchreiber, FalkStadler, Peter2019-02-202019-02-202009978-3-88579-251-2https://dl.gi.de/handle/20.500.12116/20306Graphs are often used to describe and analyze the geometry and physicochemical composition of biomolecular structures, such as chemical compounds and protein active sites. A key problem in graph-based structure analysis is to define a measure of similarity that enables a meaningful comparison of such structures. In this regard, so-called kernel functions have recently attracted a lot of attention, especially since they allow for the application of a rich repertoire of methods from the field of kernel-based machine learning. Most of the existing kernel functions on graph structures, however, have been designed for the case of unlabeled and/or unweighted graphs. Since proteins are often more naturally and more exactly represented in terms of node-labeled and edge-weighted graphs, we propose corresponding extensions of existing graph kernels. Moreover, we propose an instance of the substructure fingerprint kernel suitability for the analysis of protein binding sites. The performance of these kernels is investigated by means of an experimental study in which graph kernels are used as similarity measures in the context of classification.enGraph-kernels for the comparative analysis of protein active sitesText/Conference Paper1617-5468