A REGIONALIZABLE STATISTICAL MODEL OF INTERSECTING REGIONS IN PROTEIN{LIGAND BINDING CAVITIES

摘要

Finding elements of proteins that in°uence ligand binding speci¯city is an essential aspect ofnresearch in many ¯elds. To assist in this e®ort, this paper presents two statistical models, basednon the same theoretical foundation, for evaluating structural similarity among binding cavities.nThe ¯rst model specializes in the uni¯ed" comparison of whole cavities, enabling the selectionnof cavities that are too dissimilar to have similar binding speci¯city. The second model enables anregionalized" comparison of cavities within a user-de¯ned region, enabling the selection ofncavities that are too dissimilar to bind the same molecular fragments in the given region. Wenapplied these models to analyze the ligand binding cavities of the serine protease and enolasensuperfamilies. Next, we observed that our uni¯ed model correctly separated sets of cavities withnidentical binding preferences from other sets with varying binding preferences, and that ournregionalized model correctly distinguished cavity regions that are too dissimilar to bind similarnmolecular fragments in the user-de¯ned region. These observations point to applications ofnstatistical modeling that can be used to examine and, more importantly, identify in°uentialnstructural similarities within binding site structure in order to better detect in°uences onnproteinu0001ligand binding speci¯city