15:45 GPU based Acceleration of de Novo Protein Tertiary Structure Prediction – T. Ishida (Tokyo Tech)
Predicting three-dimensional structures of proteins from their amino acid sequences is one of the central challenges in computational biology. De novo predictions, which seek to build tertiary structure models from just their amino acid sequences without template structures, are much more difficult to template-based predictions and require vast computational resources. Here, we tried to accelerate a de novo protein tertiary structure prediction based on the fragment assembly method, by using GPU computing techniques. Although GPUs are designed specifically for computer graphics and thus are very limited in terms of operations and programming, they can operate massive parallel jobs much faster than CPUs because they have hundreds of streaming multi-processors in the core. In this work, we parallelized the calculation process of the potential energy function in the conformational sampling process and implemented it as a GPU kernel function. As results, the GPU-accelerated system achieved a speedup of up to 3.4 times with respect to a single CPU core.