Jens Sadowski

Comparison of Automatic Three-Dimensional Model Builders Using 639 X-ray Structures

Several criteria were defined to select a dataset of high-quality X-ray structures from the Cambridge file resulting in 639 molecules. Six currently available programs for automatic 3D structure generation were compared by converting the connectivity tables including appropriate stereodescriptors from this dataset of 639 molecular structures into 3D geometries: CONCORD, ALCOGEN, Chem-X, MOLGEO, COBRA, and CORINA. The geometries produced by the different programs were evaluated in terms of several quality criteria and are discussed in detail. These criteria measure how well the different programs reproduce the X-ray geometries of the 639 input structures. Accordingly, the major strengths and weaknesses of the programs are indicated.

Locating Biologically Active Compounds in Medium-Sized Heterogeneous Datasets by Topological Autocorrelation Vectors: Dopamine and Benzodiazepine Agonists

Electronic properties located on the atoms of a molecule such as partial atomic charges as well as electronegativity and polarizability values are encoded by an autocorrelation vector accounting for the constitution of a molecule. This encoding procedure is able to distinguish between compounds being dopamine agonists and those being benzodiazepine receptor agonists even after projection into a two-dimensional self-organizing network. The two types of compounds can still be distinguished if they are buried in a dataset of 8323 compounds of a chemical supplier catalog comprising a wide structural variety. The maps obtained by this sequence of events, calculation of empirical physicochemical effects, encoding in a topological autocorrelation vector, and projection by a self-organizing neural network, can thus be used for searching for structural similarity, and, in particular, for finding new lead structures with biological activity.

The comparison of geometric and electronic properties of molecular surfaces by neural networks: Application to the analysis of corticosteroid-binding globulin activity of steroids

SummaryIt is shown how a self-organizing neural network such as the one introduced by Kohonen can be used to analyze features of molecular surfaces, such as shape and the molecular electrostatic potential. On the one hand, two-dimensional maps of molecular surface properties can be generated and used for the comparison of a set of molecules. On the other hand, the surface geometry of one molecule can be stored in a network and this network can be used as a template for the analysis of the shape of various other molecules. The application of these techniques to a series of steroids exhibiting a range of binding activities to the corticosteroid-binding globulin receptor allows one to pinpoint the essential features necessary for biological activity.

The generation of 3D models of host-guest complexes

Abstract A program has been developed for the completely automatic generation of 3D atomic coordinates of organic molecules. This system has been extended also to process polymacrocyclic structures, allowing complex host molecules to be modelled. A three-step approach has been developed for modelling the geometry and energies of host-guest complexes which avoids the problem of being trapped in local minima. Two examples are given that illustrate that fine experimental details can be derived from this approach.

Optimization of the drug-likeness of chemical libraries

A scoring scheme for the classification of moleculesinto drugs and non-drugs was established. It was setup by using atom type descriptors for encoding themolecular structures and by training a feed-forward neural network for classifying the molecules. The approach was parameterized by using large databases of drugs and non-drugs - the Available Chemicals Directory (ACD) with 169 331 molecules and the World Drug Index (WDI) with 38 416 molecules. It was able to reveal features in the molecular descriptors that either qualify or disqualify a molecule for being a drug. The method classified about 80% of the ACD and the WDI correctly. It was extended to the application for crop protection compounds and can be used to prioritize compounds for synthesis, purchase, or biological testing. An enhancement allows to optimize the drug character of combinatorial libraries.

Evaluation of Molecular Surface Properties Using a Kohonen Neural Network

In this chapter the use of a Kohonen map in structure–activity relationship (SAR) and drug design is described. A transformation of 3D-molecular surfaces into 2D-Kohonen maps is realized on the basis of an application of a Kohonen neural network. The trained neurons of the Kohonen maps are then colored according to the molecular electrostatic potential (MEP) values on the van der Waals surfaces. A template approach is presented for the comparison of the shape and the MEP values of a molecule with that of a reference structure. Applications of these methods are illustrated with datasets of ryanodines, cardiac glycosides, and steroids. The results indicate that the Kohonen neural network may be a useful tool for the investigation of large datasets of molecules and for the fast and accurate comparison of molecular electrostatic potentials and shapes within a series of compounds.

3D Structure Descriptors for Biological Activity

Novel ways of coding the structure of chemical compounds are presented and their use for correlating biological activity is explored. These structure codes take account of the three-dimensional arrangement of the atoms in a molecule, or consider molecular surface properties. These molecular representations have been studied with large datasets; various applications to biological activity studies and the definition of chemical diversity will be presented.

A Systemsatic Approach to Finding New Lead Structures Having Biological Activity

The development of a new drug is an enormously largescale and expensive process. Thus, computer simulation methods become to play an increasing role in the development of new pharmacologically active compounds. Most of the commercial software presently used, comes from the U.S.; their deficits have become more and more obvious during the last years. Several methods have been developed in our project to alleviate these problems. The search for new lead structures starts with analyzing large databases of compounds (several hundreds of thousands up to several millions of compounds) zeroing into a few promising structures by increasing sophistication of structure representation. Due to the large number of chemical compounds, a systematic scheme for representing structures was developed: The starting point is the constitution, followed by calculation of the 3D structure, then including conformational flexibility. At each step, a variety of chemical properties can be taken into consideration. In addition, new programs have been developed for the treatment of conformational flexibility. The methods presented are also useful for other areas of application dealing with chemical information. Thus, it was shown that one of these new structure representations is suitable for treating problems in combinatorial synthesis. Neural networks and genetic algorithms are highly important for the investigation of the correlation between structure and biological activity. Complex relationships and huge amounts of data can be processed by these methods. Implementation of these procedures on highly parallel computers has proved that datasets of several hundreds of thousands of structures can be treated with acceptable computation times.