Techniques for the calculation of three-dimensional structural similarity using inter-atomic distances

Abstract

This paper reports a comparison of several methods for measuring the degree of similarity between pairs of 3-D chemical structures that are represented by inter-atomic distance matrices. The methods that have been tested use the distance information in very different ways and have very different computational requirements. Experiments with 10 small datasets, for which both structural and biological activity data are available, suggest that the most cost-effective technique is based on a mapping procedure that tries to match pairs of atoms, one from each of the molecules that are being compared, that have neighbouring atoms at approximately the same distances.

Chemical Similarity Searching

“…Willett and co-workers have described both distancebased 52 and angle-based 53 descriptors for the calculation of 3D similarity. The simplest of the distance-based descriptors is the distance distribution in which each distance in a molecule increments a count in an associated distance-range bin.…”

Chemoinformatics Research at the University of Sheffield: A History and Citation Analysis

“…An example of the nearest neighbours obtained in a Tanimoto-based similarity search is shown in Figure 5, where the close relationship to the target structure is clearly evident. Later work in Sheffield [64] evaluated the inter-atomic distance screens that are used for 3D substructure searching when applied to the calculation of 3D structural similarities, and this occasioned several subsequent discussions of the use of both distance-based and angle-based methods for 3D similarity searching (see, e.g., [65][66][67]). However, most of the research, both in Sheffield and elsewhere, of 3D similarity measures has focused on two alternative approaches: the use of the maximum common substructure (MCS) and of molecular field overlaps.…”

Effectiveness of retrieval in similarity searches of chemical databases: a review of performance measures

“…For example, the Sheffield group has generally quoted the mean numbers of active compounds identified in some number (e.g., the top-20) of the nearest neighbours, when averaged over a set of searches for bioactive target structures; an example is a study of distance-based measures for 3D similarity searching [21]. Alternatively, the Merck group have used cumulative recall diagrams, from which it is simple to obtain the enrichment, i.e., the number of actives retrieved relative to the number that would be retrieved if compounds were picked from the database at random.…”