Automated site-directed drug design: Approaches to the formation of 3D molecular graphs

Abstract

This paper presents two methods for the automated generation of 3D molecular graphs. The objective is to obtain molecular graphs that span the binding site and incorporate predicted ligand points at their vertices. The steric surface at the site forms the mould into which a developing ligand has to fit. Patterns of molecular forces are found at the receptor's accessible surface that are important in controlling drug binding. In many cases of drug-receptor interaction, key atoms termed site points are located at the site surface. These give rise to loci in which putative ligand atoms may be non-covalently attached to the site. The drug designer has to link up the ligand points by a network of bonds to form a molecular graph for the ligand. Structure generation is a combinatorial problem that can only be solved in exponential time by brute-force methods, so heuristics must be used to obtain answers. The site-point ligand-point model can be used to indicate putative ligand points and provide spatial constraints for the evolving ligand. This model has been successfully used when filling a binding site with a minimal 2D molecular graph [1]. The goal of this work is to extend these methods into 3D through the generation of connecting chains or graphs. Once a graph has been established, appropriate atoms and bonds will have to be placed at the vertices and along the edges of the graph; if the ligand is to be recognised by the site, then the pattern of molecular forces generated by the ligand must match a complementary pattern of forces presented by the site. An automatic method of 3D structure generation would represent an important addition to the armoury of the drug designer. A drug designer would like to create a structure that can occupy empty ligand points in the binding site and join them to other ligand points, or to seed atoms, in an existing ligand. This objective can be achieved by finding the minimal molecular graph that incorporates these specified points. The graph should (i) not cut across the receptor-accessible surface, thereby causing repulsive steric interactions; (ii) match the local patterns of electrostatic and hydrophobic forces in the receptor site; (iii) be able to meet these requirements without having to adopt a high-energy conformation. The resulting subgraph should then be near an enthalpic minimum: any entropic penalty can be reduced by later addition of rings which reduce the torsional freedom of the chain. An algorithm is presented that generates molecular graphs to fulfill the steric and torsional requirements. An alternative database strategy is also discussed.