Protein structure comparison based on profiles of topological motifs
Juris Viksna1, David Gilbert2, Gilleain Torrance
1jviksna@cclu.lv, Institute of Mathematics and Computer Science, University of Latvia; 2drg@brc.dcs.gla.ac.uk, Bioinformatics Research Centre, Department of Computing Science, University of Glasgow
We report on a new approach to protein structure comparison using the existing TOPS database. Essentially TOPS comparisons are based on searching for subgraph isomorphisms and/or maximal common subgraphs. Whilst it would be very desirable to deal with subgraphs that contain also some negative information (e.g. the absence of a particular edge), doing this is computationally unrealistic. Instead we propose a comparison against a profile of "positive" and "negative" motifs, the later can be considered as a way to capture the required negative information. Such approach allows significantly increase the accuracy of predictions – we have up to 73% correct predictions, instead of 32% obtained by using only the best motif.