Nearest Neighbor Classification in 3D Protein Databases

In molecular databases, structural classification is a basic task that can be successfully approached by nearest neighbor methods. The underlying similarity models consider spatial properties such as shape and extension as well as thematic attributes. We introduce 3D shape histograms as an intuitive and powerful approach to model similarity for solid objects such as molecules. Errors of measurement, sampling, and numerical rounding may result in small displacements of atomic coordinates. These effects may be handled by using quadratic form distance functions. An efficient processing of similarity queries based on quadratic forms is supported by a filter-refinement architecture. Experiments on our 3D protein database demonstrate the high classification accuracy of more than 90% and the good performance of the technique.