Method to accelerate prediction of membrane protein types by CUDA

This paper introduce a parallel computing method to improve the efficiency of prediction of membrane protein types by SVM. With early hardware limitations of the GPU (lack of synchronization primitives and limited memory caching mechanisms) can make GPU-based computation inefficient. We present this efficient method for prediction of membrane protein type for Intel(R) Core(TM) i3-3110m quad-core and NVIDIA GeForce GTX610m GPU (kepler architecture). CUDA implementation improves the efficiency of the prediction by a SVM compared to a standard CPU based implementation, the study show the fact that the method is more than 65 times than that of CPU serial implementation and accuracy rate can reach 80 percent. Simultaneously, and the parallel algorithm applied to prediction of membrane protein types by SVM is an efficient approach to high performance bioengineering applications.