Most existing network motif discovery algorithms detect exact motifs.Because biological networks are incomplete and noise

and the life process is dynamic

probability network motifs are of great importance.In this paper

we propose a non-treelike network motif discovery algorithm

which focuses on network motifs derived from families of mutually similar but not necessarily identical patterns.In our approach

we first present a subgraph mining algorithm ESN for searching all non-treelike subgraphs of a given size.Then multiple graph alignment is performed on a set of subgraphs with a same size.Finally

a simulated annealing algorithm is used to derive network motifs

based on a statistical model and its corresponding scoring function.Experimental evaluation on transcriptional regulatory networks of E.coli and Yeast shows that our algorithm achieves good performance.