Heuristic Algorithms in Bioinformatics

Many bioinformatics problems are very hard to solve optimally and within polynomial time of their size. This motivates the use of heuristics in bioinformatics. The present thesis constitutes research into the design and analysis of efficient heuristic algorithms of practical usage for solving biological problems. The research made in this thesis can be related to three main objectives: 1) The design efficient tools for solving relevant bioinformatics problems, 2) The exploration of stochastic search algorithms, and 3) The use of bioinformatics tools get an increased knowledge about the world from a bioinformatics perspective. The research relates to two bioinformatics problems: Comparison of 3D protein structures also known as structural comparison, and physical mapping, which is a combinatorial problem that originates from the determination of whole genome sequences. The thesis is structured in two parts: Part I, in which the background is introduced for the algorithms used and the problem domains investigated. Part II, in which the four papers I have produced are presented. One paper has been accepted for publication in the Journal of Computational Biology, two have been published in the proceedings of the Congress on Evolutionary Computation, and one is currently in submission. The research on structural comparison resulted in a journal paper, written with William R. Taylor, and a tool, called FASE (Flexible structural Alignment from Secondary structure), for comparison, superposition, and alignment of protein structures. The chosen focus for structural comparison is reflected in the features of FASE: FASE can find non-sequential structure similarities with a novel, local application dynamic programming, FASE uses a new idea to search the space of structural alignments by making initial alignments of the protein structures according to their SSE definitions. This improves the runtime, while still enabling discovery of very subtle and remote similarities. A novel consensus scoring method, combining five measures for structural similarity, was introduced in order to increase robustness in similarity assessment. The significance of the found similarity values was determined using the SCOP classification as reference. In conclusion, FASE was shown to be robust with defaults parameters, and have a better time-complexity than competitive methods. FASE has been tested on