Computer design of potentially bioactive molecules by geometric searching with ALADDIN

Design of chemical libraries with potentially bioactive molecules applying a maximum common substructure concept

Success in small molecule screening relies heavily on the preselection of compounds. Here, we present a strategy for the enrichment of chemical libraries with potentially bioactive compounds integrating the collected knowledge of medicinal chemistry. Employing a genetic algorithm, substructures typically occurring in bioactive compounds were identified using the World Drug Index. Availability of compounds containing the selected substructures was analysed in vendor libraries, and the substructure-specific sublibraries were assembled. Compounds containing reactive, undesired functional groups were omitted. Using a diversity filter for both physico-chemical properties and the substructure composition, the compounds of all the sublibraries were ranked. Accordingly, a screening collection of 16,671 compounds was selected. Diversity and chemical space coverage of the collection indicate that it is highly diverse and well-placed in the chemical space spanned by bioactive compounds. Furthermore, secondary assay-validated hits presented in this study show the practical relevance of our library design strategy.

SPROUT: a program for structure generation

SPROUT is a new computer program for constrained structure generation that is designed to generate molecules for a range of applications in molecular recognition. It uses artificial intelligence techniques to moderate the combinatorial explosion that is inherent in structure generation. The program is presented here for the design of enzyme inhibitors. Structure generation is divided into two phases: (i) primary structure generation to produce molecular graphs to fit the steric constraints; and (ii) secondary structure generation which is the process of introducing appropriate functionality to the graphs to produce molecules that satisfy the secondary constraints, e.g., electrostatics and hydrophobicity. Primary structure generation has been tested on two enzyme receptor sites; the p-amidino-phenyl-pyruvate binding site of trypsin and the acetyl pepstatin binding site of HIV-1 protease. The program successfully generates structures that resemble known substrates and, more importantly, the predictive power of the program has been demonstrated by its ability to suggest novel structures.

Pharmacophoric pattern matching in files of three-dimensional chemical structures: characterization and use of generalized torsion angle screens

This paper describes the use of generalized torsion angles for the screening of conformational searches in databases of three-dimensional chemical structures. A generalized torsion angle is defined as the dihedral angle between two vectors, A1-A2 and A3-A4, in which none, some, or all of the vectors A1-A2, A2-A3, and A3-A4 correspond to formal chemical bonds. The screens consist of a set of four atoms together with an associated angular range, and are identified by a statistical analysis of the frequencies of occurrence of these features in the Cambridge Structural Database. These frequencies are discussed, and the effectiveness of the screens is demonstrated by an extensive series of searches for representative pharmacophoric patterns.

GenStar: a method for de novo drug design

A novel method, which we call GenStar, has been developed to suggest chemically reasonable structures which fill the active sites of enzymes. The proposed molecules provide good steric contact with the enzyme and exist in low-energy conformations. These structures are composed entirely of sp3 carbons which are grown sequentially, but which can also branch or form rings. User-selected enzyme seed atoms may be used to determine the area in which structure generation begins. Alternatively, GenStar may begin with a predocked 'inhibitor core' from which atoms are grown. For each new atom generated by the program, several hundred candidate positions representing a range of reasonable bond lengths, bond angles, and torsion angles are considered. Each of these candidates is scored, based on a simple enzyme contact model. The selected position is chosen at random from among the highest scoring cases. Duplicate structures may be removed using a variety of criteria. The compounds may be energy minimized and displayed using standard modeling programs. Also, it is possible to analyze the collection of all structures created by GenStar and locate binding motifs for common fragments such as benzene and naphthylene. Tests of the method using HIV protease, FK506 binding protein (FKBP-12) and human carbonic anhydrase (HCA-II) demonstrated that structures similar to known potent inhibitors may be generated with GenStar.

Pharmacophoric pattern matching in files of three-dimensional chemical structures: use of bounded distance matrices for the representation and searching of conformationally flexible molecules

This paper discusses the use of bounded distance matrices for the representation of conformationally flexible three-dimensional (3D) molecules. It is shown that pharmacophoric pattern searches of databases of flexible 3D molecules represented in this way can be carried out using screen and geometric searching algorithms that are analogous to those used for searching databases of rigid 3D structures. Molecules matching a query pattern after the geometric search must then undergo a final conformational search to determine whether they can, in fact, adopt a conformation that matches the query. An analysis of this three-stage searching procedure shows that searching databases of flexible 3D molecules is extremely demanding of computational resources.

Pharmacophoric pattern matching in files of three-dimensional chemical structures: characterization and use of generalized valence angle screens

This paper describes the use of generalized valence angles for the screening of pharmacophoric pattern searches in databases of three-dimensional chemical structures. A generalized valence angle is defined as the angle between two vectors, AB and BC, which have a common vertex B, and in which both vectors correspond to formal chemical bonds; one vector corresponds to a bond and the other to a non-bonded interaction; or both vectors correspond to non-bonded interactions. The screens are identified by a statistical analysis of the frequencies of occurrence of these angle-based features in the Cambridge Structural Database. The occurrence frequencies are discussed and shown to be explicable in terms of small, commonly occurring structural features. The effectiveness of the screens is demonstrated by an extensive series of searches for representative pharmacophoric patterns. The results are compared with those obtained from a similar series of searches using distance-based screens: The latter are found to give a better level of performance, and evidence is presented to suggest that this is due to a high degree of association between the assignments of the angle-based screens.