A Fast and Memory-Efficient Hierarchical Graph Clustering Algorithm

Abstract

In this paper we propose a quick and memory-efficient implementation of the TRIBE-MCL clustering algorithm, suitable for accurate classification of large-scale protein sequence data sets. A symmetric sparse matrix structure is introduced that can efficiently handle most operations of the main loop. The reduction of memory requirements is achieved by regrouping the operations performed during the expansion matrix squaring. The proposed algorithm is tested on synthetic protein sequence data sets of up to 250 thousand items. The validation process revealed that the proposed method performs in 30% less time than previous efficient Markov clustering algorithms, without losing anything from the partition quality. This novel implementation makes it possible for the user of an ordinary PC to process protein sequences sets of 100,000 items in reasonable time.