The Use of Artificial Intelligence Techniques for Protein Structure Prediction

Ismail, Roslan Bin (1989) The Use of Artificial Intelligence Techniques for Protein Structure Prediction. PhD thesis, University of Glasgow.

Abstract

The conventional technique for computerized protein structure prediction uses several programming languages such as Fortran, C, Pascal etc. With recent advances in programming languages and the development of rule-based systems, the computerized part of the problem is undergoing major change. This thesis sets out the idea of extending the properties of an intelligent rule-based system and recognising incomplete nature of knowledge for this problem. It reviews the existing architectures and characteristics that embody an intelligent system. As the outcome of the idea, a new system called PREDMOLL, written in Prolog, is developed. PREDMOLL is based on the blackboard architecture with several other extra features. This thesis also reviews some current uncertainty techniques and developes a formula based on a modifications of the Bayes theorem, to deal with multiple hypotheses. The problem of conditional independence assumption is reduced to the minimum. The formula is used as a decision-making criterion to determine secondary structure boundaries. For tertiary structure prediction, this thesis suggests a similarity value for primary sequence homology to overcome the problem of arbitrary uncertainty values in rules. PREDMOLL and the uncertainty techniques incorporated with it are used to test the hypothesis that the performance of protein structure prediction is improved by combining several methods. The test is carried out by a series of experimental predictions with user-defined rules and predefined constraints. The behaviour of PREDMOLL during the problemsolving process of the experiments is shown. The results obtained yield improvements in precision for secondary structure prediction and further improvements are expected. For tertiary structure prediction, some preliminary progress is shown and, due to lack of genuine rules, ad-hoc rules are generated from the protein data base. The status of PREDMOLL and its advantages over other systems is discussed. Several suggestions are made to improve current facilities in PREDMOLL and problems in a wider domain. Suggestions are also made for further improvements in tertiary structure prediction.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Keywords: Computer science, Bioinformatics, Artificial intelligence
• Date of Award: 1989
• Depositing User: Enlighten Team
• Unique ID: glathesis:1989-77824
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 14 Jan 2020 11:53
• Last Modified: 14 Jan 2020 11:53
• URI: https://theses.gla.ac.uk/id/eprint/77824