Hybrid parallel metaheuristics for molecular docking on computational grids

The thesis proposes an extensive analysis of adaptive hierarchical parallel metaheuristics for ab initio conformational sampling. Standing as an NP, combinatorial, highly multi-modal optimization problem, conformational sampling requires for high-performance large scale hybrid approaches to be constructed. Following an incremental definition, minimum complexity conformational sampling mathematical models are first analyzed, entailing a review of different force field formulations. A comprehensive analysis is conducted on a large set of operators and local search algorithms including adaptive and dynamic mechanisms. As determined by the analysis outcomes, complex a priori and online parameter tuning stages are designed. finally, highly scalable hierarchical hybrid distributed algorithm designs are proposed. Experimentation is carried over multiple parallelization models with afferent cooperation topologies. Expenmentations resulted in unprecedented results to be obtained. Multiple perfect conformational matches have been determined, on highly difficult protein structure prediction and molecular docking benchmarks, with RMSD average values below 1.0A. The validation of the proposed hybrid approaehes was performed on Grid'5000, a French computational grid, with almost 5000 computational cores. A Globus Toolkit hased Grid'SOOO system image has been developed, sustaining large scale distributed deployments. The constructed hierarchical hybrid distributed algorithm has been deployed on multiple clusters, with almost 1000 computing cores. Finally, a parallel AutoDock version was developed using the ParadisEO framework, integrating the developed algorithms.