Protein domain decomposition using spectral graph partitioning

Proteins usually consist of several structural domains which play an important role in the classification and analysis of them. As the number of known protein structures grows at an exponential rate, a computer-aided domain decomposition of proteins becomes desirable. We present three different decomposition methods in this report. They are based on a graph representation of the proteins. These graphs are recursively decomposed into domains using the Fiedler vector. Our methods correctly decompose 89.1% and 90.9% of the 55 proteins in our test data set. We also show the problems with the incorrectly decomposed proteins in a detailed analysis.