PATRICIA trie based time and memory optimization for fast network motif Search

Abstract

Network motif search is useful in uncovering the important functional components of complex networks in biological, chemical, social and other domains. PATCOMP - a PARTICIA based novel approach for network motif search is proposed in this paper. The algorithm of PATCOMP takes benefit of memory compression and speed of PATRICIA trie to store the collection of subgraphs in memory and search them for classification and census of network. The structure of trie nodes and how data structure is developed to use it for counting the subgraphs is also described. PATCOMP was compared with QuateXelero and G-Tries.The main benefit of this approach is significant reduction in memory space requirement particularly for large network motifs with acceptable time performance. The experiments with directed networks like E.coli, yeast, social and electronic validated the advantage of PATCOMP in terms of reduction in memory usage by 2.7-27.7% as compared to QuateXelero for smaller motif sizes (with exceptions of s=6 for E. coli and s=6 for social), and 7.8-38.35% for larger motif sizes. For undirected networks, PATCOMP utilizes less memory by 0.07%-43% (with exception of s=7 for electronic and s=6,8 for dolphin networks).