GPU Accelerated Finding of Channels and Tunnels for a Protein Molecule

This paper proposes a novel method for computing the cavities and channels/tunnels in a protein molecule in interactive time without significant user effort. A sphere tree structure is used to represent a protein molecule, which provides a parallel architecture to access a Graphic Processing Unit (GPU) memory. The use of CUDA programming with a GPU then allows the proposed system to work in parallel on either a sphere tree structure of a molecule or a set of voxels composing the space. A real-time performance is achieved for proximity queries on a protein molecule, and an interactive time performance is realized for finding all the cavities and channel/tunnels without user effort. The proposed system also provides a method for approximating a convex hull of a molecule in a discrete space, and then generates the shortest path from a user selected or automatically chosen cavity to the exterior of the protein molecule. Experimental results in comparison with previous methods confirm the time efficiency of the proposed system.