Orthosimilarity

Because of the interdependence of descriptors, the traditional approach to estimate the similarity for a set of structures is biased. We describe use of orthogonalized descriptors for determining the degree of similarity among molecules. The approach is illustrated for octane isomers when path numbers are used as molecular descriptors. Use of orthogonalized descriptors removed the high degree of the degeneracy that characterized the similarity/dissimilarity table for octane isomers. The method outlined in the article is general. It allows one to eliminate the inherent bias caused by the interdependence of molecular descriptors (mathematical or not). The approach equally applies to compounds viewed as three-dimensional objects characterized by invariants based on three-dimensional molecular data.