Graph alignments: A new concept to detect conserved regions in protein active sites

Abstract

We introduce the novel concept of graph alignment, a generalization of graph isomorphism that is motivated by the commonly used multiple sequence alignments. Graph alignments and graph isomorphisms are equivalent in the case of pairwise comparisons, but if many graphs should be analyzed simultaneously, graph alignments are more robust against noise and perturbations. Thus, graph alignments can detect conserved patterns in large sets of related graphs in a more reliable way. We study simple heuristics for the efficient calculation of approximate graph alignments and apply our approach to the structural analysis and functional classification of the active sites of proteins. Our approach is able to automatically detect even weak- ly conserved structural patterns in protein structures. Among other applications, these motifs can be used to classify proteins according to their function.