Introduction to similarity searching in chemistry

Abstract

The similarity concept and its database implementation - similarity searching, are overviewed in the context of chemoinformatics. Similarity is defined in terms of matches/overlap, dissimilarity in terms of mismatches/difference, for qualitative/quantitative characteristics. Similarity, dissimilarity and composite measures are constructed from similarity or/and dissimilarity components. Asymmetric measures are constructed by unequal weighting of dissimilarity components. Whole objects or local regions of them are compared, yielding global or local similarity. Asymmetric local similarity is obtained by treating the objects in the comparison unequally, e.g. by ignoring parts of them. Global characteristics provide overall descriptions of objects, local characteristics provide sufficient locational information for object alignment/superposition to be effected. Similar objects are likely to have similar properties - similar property principle. In chemical similarity searching, molecules, fragments of molecules, reactions, mixtures, journal articles, etc. are selected as objects of interest. The selection of characteristics and their encoding is illustrated using the atom pair and topological torsion descriptors, as well as their variants of increased fuzziness. Similarity measure selection is still very much a matter of trial and error. Standard query object specification is made easier by using query by example, multiple searches using a single query yield a highly informative hyperlinked screen, and joint queries involve more than one object. Similarity scores illustrate results from similarity searches and measures of their effectiveness. Areas of application include direct and reverse property prediction, data mining, virtual screening, diversity analysis, pharmacophore searching, ligand docking, structure elucidation, pattern matching, and signature analysis.