Enhanced GPU accelerated K-means algorithm for gene clustering based on a merging thread strategy

Abstract

Past research has demonstrated that gene clustering can be effectively accomplished with the K-means algorithm. Furthermore, the clustering process can be conducted swiftly with the use of graphic processing units (GPUs). However, due to the limited number of processing units (cores) on a GPU, the computation time will be lengthened if the number of gene data is too large. To alleviate this problem, a novel method for realizing the K-means algorithm on GPUs has been developed and presented in this paper. Essentially, a fragment shader program is implemented to process multiple data points in a single thread. Experimental results show that our proposed GPU accelerated scheme can attain over 25% increase in the computation speed as compared with the existing method. © Springer-Verlag 2013.