Detecting Motifs from Sequences. Proceedings of the International Conf. on Machine Learning
SB Sandmeyer, Yan Hu, Dennis F. Kibler · eScholarship (California Digital Library) · 1999
The problem of multiple global comparisons of families of biological sequences has been well-studied.Fewer algorithms have been developed for identifying local consensus pa~ terns or motifs in biological sequences.These two important problems have different biological constraints and, consequently, different computational approaches.The difficulty of finding the biologically meaningful motifs results from the variability in (1) the bases at each position in t he motif, (2) the location of the motif in the sequence and (3) the multiplicity of motif occurrences within a given sequence.In addition the short length of many biologically significant motifs and the fact that motifs gain biological significance only in combinations, makes them difficult to determine using standard statistical methods.In this paper we introduce our own approach, DMS, which combines multiple o~ jective fun ctions with an improved iterative sampling search method.We compare the • main approaches for finding motifs and test the effectiveness of the various algorithms by comparing them on ten real domains and fourteen artificial domains.The main advantage of DMS is that it is better able to find shorter motifs.