相互エントロピーを用いたアライメントの改良

概要

論文の詳細を見る
We improve the algorithm to align amino acid sequences of identical protein which is one of the most fundamental operations studying the analysis of genome. In pair-wise alignment, one chooses one aligned pair (i.e., two sequences) without special reasons from several aligned pairs (the number of these pairs is often very large) giving the same smallest values to the difference properly defined between two sequences. In this paper, we compute the mutual entropy for several such pairs having the same difference, and we classify the pairs into some groups such that the same group consists of the pairs having the same value of the mutual entropy, then we finally compute the mean value of the mutual entropy over the whole groups. As a consequence, we can observe the following interesting fact for some proteins that the aligned pair obtained by usual alignment with geometrical protein structure (we call such a alignment the biological alignment here) is in the group having the value of the mutual entropy closest to the mean value of the mutual entropy. From the above observation we conclude that our method using the alignment (MOU-alignment) and the mutual entropy makes us possible to find the biological alignment, that is, we do not need to know the geometrical structure to obtain the biological alignment.
情報知識学会の論文
2003-05-24