Fred Hutchinson Cancer Research Center, 1124 Columbia St. Seattle, WA 98104 USA
A general searching method for comparing multiple sequence alignments was developed to detect sequence relationships between conserved protein regions. Multiple alignments are treated as sequences of amino acid distributions and aligned by comparing pairs of such distributions. The method is sensitive, detecting weak sequence relationships between protein families, and selective, having a low rate of false positives. Similarities can be detected beyond the range of conventional sequence database searches. For example, searching the Blocks Database against itself identified similarities between flavoprotein subunits of two oxidoreductase families and between helix-turn-helix DNA binding domains from diverse proteins. Structural and functional predictions are possible for many conserved protein regions and the results can direct various lines of experimental research. The method has been fully implemented for public use on the WWW.