hiCLUMP : A hybrid Implementation of the CLUMP Algorithm for Clustering Microarrays Data

Microarrays technology allows us to measure the expression level of hundreds of thousands of genes simultaneously. The microarrays data analysis process involves various heavy computational tasks such as clustering. The clustering can be defined as partitioning a dataset into groups where objects in the same group are similar in somehow. CLUMP (clustering through MST in parallel) is one of the minimum spanning tree (MST) -based clustering techniques. It employed a parallel approach to reduce the MST construction time. An enhanced version of CLUMP (iCLUMP) was proposed to further improve the MST construction phase using cover tree data structure. Despite that modification, the MST construction phase is still a bottleneck since it is a time consuming task. Both CLUMP and iCLUMP are based on a distributed parallel computing model. Therefore, the objective of this paper is to study a different approach of enhancement using a hybrid parallel model. The proposed algorithm; hiCLUMP (hybrid CLUMP), considers utilizing multithreading on some of the distributed partitions suggested by the CLUMP algorithm. The experimental results on six different microarrays datasets show that the load balancing strategy used in hiCLUMP succeeded to decrease the MST construction in a range between 8% and 17% on 36 processing node. Moreover, the results showed that the hiCLUMP could not outperform the iCLUMP emphasizing that using another data structure is more effective than increasing the processing power of the underlying parallel machine.