Measuring Sequence Similarity Trough Many-to-Many Frequent Correlations

Comparing pairs of sequences is a problem emerging in several application areas (ranging from molecular biology, to signal processing, text retrieval, and intrusion detection, just to cite a few) and important results have been achieved through the years. In fact, most of the algorithms in the literature rely on the assumption that matching symbols (or at least a substitution schema among them) are known in advance. This paper opens the way to a more involved mechanism for sequence comparison, where determining the best substitution schema is also part of the matching problem. The basic idea is that any symbol of one sequence can be correlated with many symbols of the other sequence, provided each correlation frequently occurs over the various positions. The approach fits a variety of problems difficult to be handled with classical techniques, particularly where strings to be matched are defined over different alphabets.