Tetrahedron Computer Methodology最新文献

SECS (Simulation and Evaluation of Chemical Synthesis) is a retrosynthetic organic synthesis design program. Building a large knowledge base is required in order to solve important synthesis problems in industry. We developed a method that automatically builds a large ALCHEM transform library. Fifteen thousand reactions have been built. These were modified and reorganized into a form usable by SECS. At the same time, we added an additional strategy generation function to the SECS strategy module. We present our method and an example of a SECS analysis of a target molecule using our large ALCHEM knowledge base.

A computer system for rational molecular design is described. The system, named as ANALOGS, provides the triangular elements consisting of functional atoms for the comparison of a number of stereochemically different compounds. Application of the system for aldose reductase inhibitors is described. The results obtained give some interesting information on the inhibitor-binding site of aldose reductase and on the structural features of the inhibitors.

This paper overviews the design and current status of the mechem system of programs, which automates partially the elucidation of chemical reaction pathways of up to moderate complexity. The system addresses most aspects of pathway elucidation, with special attention to the formation of initial hypotheses. No chemical databases are used; the sources of power instead are first, novel algorithms for reasoning about pathways, and second, experimental data on the reaction to be studied. The performance of the system is illustrated on data from the liquid-phase oxidation of propane. Some by-products of the system design have contributed to chemistry knowledge. The file mechem.lsp is included on disk in this issue as a sample of the lisp functions used in mechem.

In the context of automatic knowledge acquisition for computer-assisted synthesis planning, this paper presents an efficient algorithm for the identification of the maximal common substructures between reaction graphs. The terms of the problem are first completely specified. Then, the method used to solve it is presented and developed step by step. The formal algorithm is proposed as an appendix.

Computer-based lead finding algorithms which attempt to design, on a de novo basis, ligands that will complement a known receptor site cavity face some major problems in terms of a combinatorial design space and the synthesizability of the designed molecules. On the other hand, typical 3D database search methods provide a different set of challenges. Both of these approaches are ultimately pointed toward the same goal and can be used together productively. In this article we describe advances in both areas: we first describe extensions to our de novo ligand design software which combines (a) a tree-based conformational search over a library of fragments, and (b) a form of simulated annealing which allows designed ligands to crawl around the binding site even as their structures are changing. In the second part, we then discuss an implementation of the database approach which allows users to formulate 3D substructure, superstructure, or similarity queries based upon demonstrated or hypothetical requirements for activity. Finally, we draw the two approaches together with an example of current research interest, showing how one method can feed the other.

The process of building large three-dimensional (3D) structural databases of commercial quality presents several unique challenges. The concerns which must be faced include database size, method of 3D structure generation, batch implementation of modeling and registration, correspondence with two-dimensional structures, and quality assurance of the resulting 3D structures. We describe procedures which were developed to build two databases with CONCORD: MACCS-II Drug Data Report-3D (MDDR-3D) and Fine Chemicals Directory-3D (FCD-3D). These procedures were developed to overcome limitations in the modeling process and to increase the efficiency and reliability of registration. This paper describes these programs and techniques, and reports on their performance in building large 3D structural databases.

Structurally novel potential dopamine agonists were designed by searching databases of 3D structures to find templates that match geometric criteria and can be modified into molecules suggested for synthesis. A search of 54,296 3D structures from three different databases generated 499 structurally unique molecules that meet our geometric criteria for D-2 dopaminergic activity. The search identified 8 of 9 classes of known fused ring phenolic dopaminergic compounds and 62 other classes of fused ring compounds with potential activity. The low observed frequency of finding the same ring class more than once suggests that additional searches will design many additional molecules. Compound design based on 3D substructure searching methods appears to be equally applicable to suggesting new classes of compounds for beginning or for mature medicinal chemistry investigations and does not require the construction of special libraries of templates.

This paper describes algorithms that select a set of fragment screens for chemical substructure searching. The algorithms take as input an alphanumerically ordered list of fragments, together with their frequencies of occurrence, and produce as output a partition of this list, each portion of which contains approximately the same number of fragment occurrences. The algorithms have been developed as part of an ongoing project to develop techniques for angle-based substructure searching in files of 3-D chemical molecules; however, they are applicable to any situation requiring the selection of a set of approximately equifrequently occurring descriptors. A Pascal implementation of one of the algorithms is included on disk as SCREENS.PAS.

A general method for the prediction of organic reactions by a backpropagation neural network is described. Neural networks trained using modified Dugundji-Ugi BE-matrix representations gave excellent predictions of the regiochemistry for three different types of reactions: Markovnikov addition to alkenes, Diels-Alder and retro-Diels-Alder reactions, and Saytzeff elimination. The networks were able to extract reactivity information from examples of the reactions to develop an internal representation of the reactions without explicitly incorporating rules into the network. Since the neural network was better at interpolating than extrapolating, it is important that the training set span the set of possible reactions. The method of representation used is sufficiently general to handle most classes of organic reactions.

A new protein folding model for obtaining low-level structure information from sequence is constructed. Its form is related both to a parameterized energy function to represent the folding problem and a feed-back “neural network”. The values of unknown physical quantities appear as free parameters in this potential function. Ideas from the study of neural network models are used to develop a learning algorithm that finds values for the free parameters by using the database of known protein structures. This algorithm can be implemented in parallel on a multicomputer. The ideas are illustrated on a simple model of α-helix formation and prediction and used to investigate the role of hydrophobic forces in stabilizing helix hydrogen bonds.