"In order to make the structural comparison of protein binding sites more
efficient, we propose a two-step method that combines advantages from
graph-based clique-detection and geometric hashing-techniques. The search
for protein similarity is completely independent of sequence and fold
information. Instead, it it based on a recent approach for the automatic
extraction of binding sites from protein structures and the representation
of their geometric and physicochemical properties. We also present some
empirical results for similarity search in a medium-sized dataset."