摘要：Background: In order to improve the defects of falling into local optimum prematurely and the low global search capability in K-means algorithm for gene clustering analysis, an efficient hybrid algorithm by combining PK-means, Cellular Automata (CA) and Firefly Algorithm (FA), called PK-CA-FA is presented. Materials and Methods: In the algorithm, CA is introduced for relieving the problem of easy to fall into a local optimum at the first iterative stage of the PK-means and then FA is introduced to enhance the global search ability at the second iterative stage. Furthermore, in order to improve the computational efficiency, this algorithm is implemented on Graphics Processing Unit (GPU) with a Compute Unified Device Architecture (CUDA) parallelly. Results: For verifying its performance, the algorithm and its parallel version are utilized to cluster gene expression data on several benchmark datasets. The experimental results show that the proposed algorithm can effectively avoid being trapped in a bad local optimum and is generally more accurate and stable than PK-means algorithm. At the same time, the parallel implementation of the algorithm on GPU is significant, by which a considerable acceleration ratio with respect to CPU is obtained. Conclusion: It is concluded that the PK-CA-FA is an efficient algorithm for gene clustering with strong accuracy, stability and high speedup and the algorithm can be expected to find its further applications for practical gene clustering analysis.
