An information theory based multiple alignment ("Malign") method was used to align the DNA binding sequences of the OxyR and Fis proteins, whose sequence conservation is so spread out that it is difficult to identify the sites. In the algorithm described here, the information content of the sequences is used as a unique global criterion for the quality of the alignment. The algorithm uses look-up tables to avoid recalculating computationally expensive functions such as the logarithm. Because there are no arbitrary constants and because the results are reported in absolute units (bits), the best alignment can be chosen without ambiguity. Starting from randomly selected alignments, a hill-climbing algorithm can track through the immense space of s(n) combinations where s is the number of sequences and n is the number of positions possible for each sequence. Instead of producing a single alignment, the algorithm is fast enough that one can afford to use many start points and to classify the solutions. Good convergence is indicated by the presence of a single well-populated solution class having higher information content than other classes. The existence of several distinct classes for the Fis protein indicates that those binding sites have self-similar features.
