An Efficient Parallel GPU Evaluation of Small Angle X-ray Scattering Profiles

The inference of protein structure from experimental data is of crucial interest in science, medicine and biotechnology. Unfortunately, high-resolution experimental methods can not yet provide a detailed analysis of the ensemble of conformations adopted under physiological conditions. Low resolution techniques are often better suited for this task. Small angle X-ray scattering (SAXS) plays a major role in investigating important biological questions regarding the structure of multidomain proteins connected by flexible linkers or the aggregation processes that underlie several major diseases in humans. In silico simulations can bridge the gap between low resolution information and models derived from highresolution techniques. For that, it is necessary to be able to calculate the low resolution information from a given detailed model using a so-called forward model. These calculations need to be performed many times during a conformational search, and therefore need to be computationally efficient. We present an efficient implementation of the forward model for SAXS experiments with full hardware utilization of General Purpose Graphical Processor Units (GPGPUs). The proposed algorithm is orders of magnitude faster than an efficient CPU implementation, and implements a caching procedure ready to be employed in the partial SAXS evaluations required by in silico simulations.