Combining sequence and structure information in protein alignments

For distantly related proteins, alignmentsbased on structural information are more reliable than traditional sequence alignments. However, when structural comparison leaves some ambiguity in alignment, sequence information can provide valuable additional information to discriminate between multiple alternatives. In this paper we present a Bayesianmodel that incorporates sequence information into structural alignments in an automatic and adaptive fashion. By use of an estimated measure of conservation between sequence and structure, we construct an ensemble sequence/structure alignment tool capable for building refined protein alignments and identifying conserveed functional regions. This model also provides a natural tool for using structural alignment to determine evolutionary distance, which may allow phylogenetic analysis of proteins at significantly larger divergence times. Two examples are presented and compared with previous analysis in the