Employing heterogeneous computer systems for molecular docking

We have developed a prototype parallel molecular docking algorithm that enables exploitation of heterogeneous computer architectures and thus the computational power of modern PCs and supercomputer clusters. We employed the OpenCL soft-ware framework that supports hardware from multiple vendors. The parallel algorithm searches for the optimal pose of molecu-les/ligands in the protein receptor site in an in silico experiment named molecular docking. We are using an empirical scoring function to evaluate solutions. The developed algorithm was tested on several protein complexes, and the results were compared with the reference implementation of the CmDock software. We analysed the quality of the resulting ligand poses and the speed of the algorithm to demonstrate a significantly faster performance on GPUs. The reported algorithm leaves space for additional optimizations and development of the scoring function to obtain the optimal ligand poses. The source will be made available at the https://gitlab.com/Jukic/cmdock/.