Fast Implementation Method Of Protein Spots Detection Based On CUDA

The extraction of differential protein spots is one of the important methods of proteomics, which is the key step of protein analysis and research in the 2D-GEL image; provide the basis for disease diagnosis, drug research and analysis of environmental pollution. And protein detection is the crux of the extraction of differential protein spots. According to the problem of low efficiency on the method of protein detection, this paper using the characteristics of the GPU parallel threads and the model of CUDA. to study on protein detection in the 2D-GEL image. The main working content and research achievements are as follows:The parallel computing architecture for high performance based on CUDA was introduced. Firstly, the structure of procedures and cooperative computing of CUDA was studied. Secondly, the knowledge of software and hardware systems in CUDA was introduced in detail. Thirdly, the model of thread base on CUDA was introduced. The last, the knowledge of memory structure by CUDA was overview. They can to be the knowledge reserve of CUDA parallel algorithm to the 2D-GEL images preprocessing and protein spots detection.The fast implementation preprocessing method of 2D-GEL images based on CUDA was studied in detail. Firstly, the parallel of NLM algorithm was used to filtering processing of gel image. Secondly, to eliminate background uneven of gel image, the parallel of top hat transform method was used to correct image background base on CUDA architecture. Thirdly, for the characteristics of gel image, this paper determine the thread and memory for CUDA architecture, and efficiently implementing the NLM and top-hat algorithm for parallel through the optimization of asynchronous processes. Finally, under the NVIDIA GTX605 GPU, and in adoption of VS2010 for building the experimental platform, the CPU serial algorithm and GPU parallel algorithm were achieved. The result turns out that the operative efficiency of the fast detection method in this paper was obviously improved compared with CPU serial operation method.The fast implementation method of protein spots detection based on CUDA was studied in detail. Firstly, according to the principle of internal and external markers watershed algorithm, the method of parallel was designed. Secondly, overlap un-separated proteins for pre-test results in a variety of overlapping protein spots were analyzed using overlapping shapes mark a watershed algorithm proteins were separated on the basis of this study parallel design ideas. Thirdly, on the characteristics of gel image and the basis of executive model of single instruction multiple threads of CUDA, separating data space into 2D image blocks, and dividing the whole algorithm into multiple sub operations. And then the various optimizing measures such as shared memory and 2D texture memory used to achieve the fast implementation method of protein spots detection. The last, under the experimental platform of VS2010, the CPU serial algorithm and GPU parallel algorithm were discussed. The result turns out that the operative efficiency of the fast detection method described in this paper was obviously improved compared with CPU serial operation method.