Protein Folding and Self-Avoiding Walks Polyhedral Studies and Solutions
Agnes Dittel
ISBN 978-3-8325-2023-6
361 pages, year of publication: 2008
price: 42.00 €
The protein folding problem refers to the correlation of a protein's amino acid sequence and its native three-dimensional structure which is essential for functionality. It still constitutes one of the major challenges in computational biology. One commonly studied model for the protein folding problem is the HP lattice model in which proteins are considered in a fairly abstract representation. However, the HP model proteins exhibit significant parallels to proteins occurring in nature. The solution of the HP lattice mode as a combinatorial optimization problem has been proven to be NP-complete, and there have already been developed various different approaches for efficient algorithms. We study an integer programming formulation of the problem. Starting with an analysis of this model, where we concentrate on symmetry issues, we show how the model can be consolidated by exploiting symmetry properties of the underlying lattice. The main focus lies in the development of specific components of a branch-and-cut framework for the computation of solutions for the HP model by means of integer programming methods. In order to understand the structure of the model, we perform a series of polyhedral studies from which we derive two main classes of cutting planes. Furthermore, we exploit the knowledge of folding principles which are also valid for HP model proteins for the development of related branching strategies. For the solution of a special class of instances, we present an implementation of a genetic algorithm for the generation of primal feasible start solutions. Finally, we document the performance of the methods developed for each of the four topics (model consolidation, primal method, branching strategy and cutting planes) within the branch-and-cut procedure. We present computational results for different types of lattices, where we both consider known benchmark instances from literature and random instances.