Protein database searches frequently can reveal biologically significant sequence relationships useful in understanding structure and function. Weak but meaningful sequence patterns can be obscured, however, by other similarities due only to chance. By searching a database for multiple as opposed to pairwise alignments, distant relationships are much more easily distinguished from background noise. Recent statistical results permit the power of this approach to be analyzed. Given a typical query sequence, an algorithm described here permits the current protein database to be searched for three-sequence alignments in less than 4 min. Such searches have revealed a variety of subtle relationships that pairwise search methods would be unable to detect.
