A min-cut algorithm for the consistency problem in multiple sequence alignment

Abstract

Motivation: Multiple sequence alignments can be constructed on the basis of pairwise local sequence similarities. This approach is rather. exible and can combine the advantages of global and local alignment methods. The restriction to pairwise alignments as building blocks, however, can lead to misalignments since weak homologies may be missed if only pairs of sequences are compared. Results: Herein, we propose a graph-theoretical approach to find local multiple sequence similarities. Starting with pairwise alignments produced by DIALIGN, we use a min-cut algorithm to find potential (partial) alignment columns that we use to construct a final multiple alignment. On real and simulated benchmark data, our approach consistently outperforms the standard version of DIALIGN where local pairwise alignments are greedily incorporated into a multiple alignment.