Constraction of a 3D feature fragment dictionary of molecules for SAR studies

Abstract

It is well known that molecular properties, including biological functions, are closely related not only to the atomic connectivity but also to the 3D spatial arrangements of the atoms. In the present work, we have developed a computer program to identify the maximal 3D common substructures between a pair of molecular structures based on a graph theoretical clique-finding algorithm. It is applied for the 3D structural similarity search for a given data set. A query molecule and each molecule in the data set are compared using the program, and the number of atoms of the identified 3D common substructure is defined as a similarity measurement. A relative measurement such as Tanimoto coefficient is also used. Then, the feature fragments are extracted from the identified substructures for an every pair of structures in the database. Each atom (or atomic group) of a reference molecule is weighted by the frequency of appearance, and selected as constitute atoms of the feature fragment. Molecular properties such as an activity class are used to construct a 3D feature fragment dictionary.