Analysis and comparision of information theory-based distances for genomic strings

Genomic string comparison via alignment are widely applied for mining and retrieval of information in biological databases. In some situation, the effectiveness of such alignment based comparison is still unclear, e.g., for sequences with non-uniform length and with significant shuffling of identical substrings. An alternative approach is the one based on information theory distances. Biological data information content is stored in very long strings of only four characters. In last ten years, several entropic measures have been proposed for genomic string analysis. Notwithstanding their individual merit and experimental validation, to the nest of our knowledge, there is no direct comparison of these different metrics. We shall present four of the most representative alignment-free distance measures, based on mutual information. Each one has a different origin and expression. Our comparison involves a sort of arrangement, to reduce different concepts to a unique formalism, so as it has been possible to construct a phylogenetic tree for each of them. The trees produced via these metrics are compared to the ones widely accepted as biologically validated. In general the results provided more evidence of the reliability of the alignment-free distance models. Also, we observe that one of the metrics appeared to be more robust than the other three. We believe that this result can be object of further researches and observations. Many of the results of experimentation, the graphics and the table are available at the following URL: http://people.na.infn.it/similar to wbalzano/BIO.