Methods for Pattern Discovery in Unaligned Biological Sequences

Pattern discovery in biological sequences is the problem of ®nding patterns that are overrepresented in a set of unaligned DNA or protein sequences of related biological function. Such patterns could correspond to regions of the sequences responsible for the function itself, and could be used later for the functional annotation of newly determined sequences. Despite many studies, this problem can be considered far from being solved. The main dif®culty lies in the fact that signi®cant patterns can appear within each sequence with mutations, insertions or deletions of nucleotides or amino acids, without losing their biological function. This paper provides a survey of a number of existing pattern discovery algorithms, focusing both on the methods underlying them and their availability for the scienti®c community. INTRODUCTION In the last few years, the amount of biological data generated by the scienti®c community has grown exponentially, and a huge number of sequences are now available in several databases. Thus, the focus of many research projects has shifted from the generation of the data to their analysis, that is, the extraction of any kind of biological meaning from the sequences. Given a set of functionally related sequences, the main aim of patterndiscovery algorithms is to ®nd new and a priori unknown patterns that appear in every sequence, or at least in a signi®cant number of sequences, of the set. Such patterns could correspond to the regions of the sequences responsible for their function, and could be also used later for the functional annotation of newly determined sequences. From a computational point of view, the main dif®culty lies in the fact that the same pattern can appear within each sequence in a different and approximate form (ie with mutations, insertions or deletions of nucleotides or amino acids), keeping intact its biological function. The longer the patterns and the more degraded their occurrences, the harder (and slower) it is to ®nd them for patterndiscovery algorithms. A large number of different methods has been so far introduced, but we do not have yet good models or reliable algorithms that guarantee to ®nd all (or most of) the biologically meaningful patterns. This paper, without the claim of being exhaustive, provides a survey of a number of different approaches to the problem, presenting the main ideas underlying the methods and mentioning, when available, the software tools based on the different algorithms.