Modelling of Protein Surface Using Parallel Heterogeneous Architectures

A proper representation of protein surfaces is an important task in bioinformatics and biophysics. In a previous work we described a parallel workflow, based on the isosurface extraction and the CUDA architecture, able to produce high-resolution molecular surfaces based on the Van der Waals, Solvent Accessible, Richards-Connolly and Blobby definitions. In particular it is able to create surfaces composed by hundred millions triangles in less than 30 s using a Nvidia GTX 580, with speedup values up to 88. However in most application such number of triangles can be difficult to manage. In this paper we present an extension able to reduce the size of the surfaces by performing a simplification step, keeping however an high quality of the results. In particular the focus of the paper is on the efficient use of heterogeneous compute capabilities available on present workstations: the large surface produced using the CUDA device is progressively transferred and simplified on the host using the multicore CPU.