Pattern-matching algorithms are a powerful tool for finding similarities and relationships among the steadily growing amount of known protein sequences. We present a fast, sensitive pattern-matching algorithm that describes a pattern by its physico-chemical properties rather than by occurrence of amino acids, using a fast, dynamic programming algorithm. Selected examples will demonstrate applications and advantages of our approach.

For the identification of newly sequenced proteins it is necessary to have a large stock of known proteins for comparison. In this paper we present an automatically generated protein sequence database. The translation program introduced allows a periodical translation of every new release of the EMBL database. Possible errors of the translation are discussed as well as the reliability of the nucleotide sequence data, which turns out to be quite good. A comparison of our translated database with some established ones is given.

A program for the multiple alignment of protein sequences is presented. The program is an extension of the fast alignment program by Wilbur et al. (1984) into higher dimensions. The use of hash procedures on fragments of the protein sequences increases the speed of calculation. Thereby we also take into account fragments which are present in some, but not in all, sequences considered. The results of some multiple alignments are given.