Applications of artificial intelligence in conformational analysis.

Abstract

Conformational analysis provides a means of understanding a wide variety of chemical interactions. However, the complexity of the potential energy hypersurface for large molecules has restricted the use of conformational search in molecular modeling. The model building, or template joining, method employed by the WIZARD program is capable of overcoming many of the shortcomings of commonly used conformational search programs. While WIZARD has been shown to be widely applicable, the program still possesses a few limitations. This dissertation describes work done to overcome these limitations. When WIZARD is used to perform a conformational search on large, flexible molecules, the number of fragment combinations becomes very large and the conformational search can be extremely time consuming. Section I of this dissertation presents WIZARD III, a new version of the WIZARD program which is capable of applying a number of different search strategies to the conformational analysis problem. By employing search techniques such as genetic algorithms and simulated annealing, WIZARD III is capable of performing extremely rapid conformational analysis on large systems. Any program which performs molecular modeling based on an internal knowledge base will be only as good as the axioms it possesses. It would be desirable to create a program which is capable of integrating new knowledge with minimal interaction from the user. Section II of this thesis presents the MOUSE program, which utilizes inductive machine learning to derive new rules of conformational analysis. These new rules can be used to augment WIZARD's knowledge base and improve its ability to predict conformations.