A heuristic algorithm for blocked multiple sequence alignment

Blocked multiple-sequence alignment (BMA) refers to the construction of multiple alignments in DNA by first aligning conserved regions into what we call "blocks" and then aligning the regions between successive blocks to form a final alignment. Instead of starting from low-order pairwise alignments, we propose a new way to form blocks by searching for closely related regions in all input sequences, allowing internal spaces in blocks as well as some degree of mismatch. We address the problem of semi-conserved patterns (patterns that do not appear in all input sequences) by introducing into the process two similarity thresholds that are adjusted dynamically according to the input. A method to control the number of blocks is also presented to deal with the situation when input sequences have so many similar regions that it becomes impractical to form blocks by trying every combination. BMA is an implementation of this approach, and our experimental results indicate that this approach is efficient, particularly on large numbers of long sequences with well-conserved regions.