Parallel Desolvation Energy Term Calculation for Blind Docking on GPU Architectures

Abstract

In the recent literature, drug design relying on molecular docking (MD) techniques is becoming a very promising field. Most of these techniques rely on the way ligands interact with protein target using only one binding site, in addition, they ignore the fact that assorted ligands interact with unconnected parts of the target. However, by taking the latter fact into consideration, the computational complexity grows exponentially, and the energy calculation to evaluate poses on each binding site becomes more intensive. These two challenges are massively parallel in nature, thus they are very suitable for the field of high performance computing (HPC). This paper proposes the design and tuning procedures of energy desolvation term for blind docking based on the NVIDIA GPU architecture. Experiments on a hundred binding sites when docking six protein-ligand complexes of different sizes, shows that our methodology achieves a good performance. Indeed, the parallel implementation gains average speedup of 225x with respect to single core, and 62x with respect to four CPU cores. This encourages the design of parallelizing processes when dealing with molecular docking techniques.