Protein-DNA binding simulation on parallel computers

The central part of this thesis is application of algorithms for analysis of binding specificity of protein-DNA complexes to complexes containing water-mediated bridges. We show that inclusion of water contacts improves the quality of prediction; examined data sets yield correct results for all base pairs that have two or more hydrogen bonds. Our prediction algorithm uses square-well, quadratic, and Lennard-Jones approximations of hydrogen bond potential in point set pattern matching, quadratic energy minimization and very fast simulated reannealing stages respectively. The following assumptions regarding protein-DNA binding are made: (1) specificity depends on direct hydrogen bonding; (2) electrostatic forces contribute strongly to stabilization and weakly to specificity; (3) most of Van der Waals and electrostatic interactions can be neglected. A data decomposition algorithm for faster evaluation of electrostatic potential of proteins in rigid body simulation is proposed. We show that it produces expected results with precision suitable for modeling of biologic systems including protein-DNA complexes. We present a parallel computer system built and used in calculations for this work.