A Multi-objective Genetic Algorithm for Peptide Optimization

The peptide-based drug design process requires the identification of a wide range of candidate molecules with specific biological, chemical and physical properties. The laboratory analysis in terms of in vitro methods for the discovery of several physiochemical properties of theoretical candidate...

Full description

Saved in:
Bibliographic Details
Main Author: Rosenthal, Susanne
Contributors: Freisleben, Bernd (Prof. Dr.) (Thesis advisor)
Format: Doctoral Thesis
Language:English
Published: Philipps-Universität Marburg 2016
Subjects:
Online Access:PDF Full Text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The peptide-based drug design process requires the identification of a wide range of candidate molecules with specific biological, chemical and physical properties. The laboratory analysis in terms of in vitro methods for the discovery of several physiochemical properties of theoretical candidate molecules is time- and cost-intensive. Hence, in silico methods are required for this purpose. Metaheuristics like evolutionary algorithms are considered to be adequate in silico methods providing good approximate solutions to the underlying multiobjective optimization problems. The general issue in this area is the design of a multi-objective evolutionary algorithm to achieve a maximum number of high-quality candidate peptides that differ in their genetic material, in a minimum number of generations. A multi-objective evolutionary algorithm as an in silico method of discovering a large number of high-quality peptides within a low number of generations for a broad class of molecular optimization problems of different dimensions is challenging, and the development of such a promising multi-objective evolutionary algorithm based on theoretical considerations is the major contribution of this thesis. The design of this algorithm is based on a qualitative landscape analysis applied on a three- and four-dimensional biochemical optimization problem. The conclusions drawn from the empirical landscape analysis of the three- and four-dimensional optimization problem result in the formulation of hypotheses regarding the types of evolutionary algorithm components which lead to an optimized search performance for the purpose of peptide optimization. Starting from the established types of variation operators and selection strategies, different variation operators and selection strategies are proposed and empirically verified on the three- and four-dimensional molecular optimization problem with regard to an optimized interaction and the identification of potential interdependences as well as a fine-tuning of the parameters. Moreover, traditional issues in the field of evolutionary algorithms such as selection pressure and the influence of multi-parent recombination are investigated.
Physical Description:287 Pages
DOI:10.17192/z2016.0862