Format

Send to

Choose Destination
See comment in PubMed Commons below
J Mol Biol. 1996 Dec 13;264(4):823-38.

Significant improvement in accuracy of multiple protein sequence alignments by iterative refinement as assessed by reference to structural alignments.

Author information

1
Department of Biochemistry, Saitama Cancer Center Research Institute, Japan.

Abstract

The relative performances of four strategies for aligning a large number of protein sequences were assessed by referring to corresponding structural alignments of 54 independent families. Multiple sequence alignment of a family was constructed by a given method from the sequences of known structures and their homologues, and the subset consisting of the sequences of known structures was extracted from the whole alignment and compared with the structural counterpart in a residue-to-residue fashion. Gap-opening and -extension penalties were optimized for each family and method. Each of the four multiple alignment methods gave significantly more accurate alignments than the conventional pairwise method. In addition, a clear difference in performance was detected among three of the four multiple alignment methods examined. The currently most popular progressive method ranked worst among the four, and the randomized iterative strategy that optimizes the sum-of-pairs score ranked next worst. The two best-performing strategies, one of which was newly developed, both pursue an optimal weighted sum-of-pairs score, where the pair weights were introduced to correct for uneven representations of subgroups in a family. The new method uses doubly nested iterations to make alignment, phylogenetic tree and pair weights mutually consistent. Most importantly, the improvement in accuracy of alignments obtained by these iterative methods over pairwise or progressive method tends to increase with decreasing average sequence identity, implying that iterative refinement is more effective for the generally difficult alignment of remotely related sequences. Four well-known amino acid substitution matrices were also tested in combination with the various methods. However, the effects of substitution matrices were found to be minor in the framework of multiple alignment, and the same order of relative performance of the alignment methods was observed with any of the matrices.

PMID:
8980688
DOI:
10.1006/jmbi.1996.0679
[Indexed for MEDLINE]
PubMed Commons home

PubMed Commons

0 comments
How to join PubMed Commons

    Supplemental Content

    Full text links

    Icon for Elsevier Science
    Loading ...
    Support Center