Abstract: MATH/CHEM/COMP 2002, Dubrovnik, June 24-29, 2002

Predicting helical transmembrane proteins from large protein databases and from genomes

Miklos Cserzo¹, Birgit Eisenhaber², Frank Eisenhaber², and Istvan Simon¹

¹Institute of Enzymology, Hungarian Academy of Sciences, P.O.Box 7., H-1518 Budapest, Hungary

²IMP Bioinformatics, Dr. Bohr-Gasse 7, A-1030 Vienna, Austria

Genome analyses need fast and efficient prediction methods. This is especially true for integral membrane proteins due to the difficulties of experimental structure determinations of these proteins. While efficient methods are available for predicting the topology of helical transmembrane proteins, these methods also report "transmembrane segments" in many non-transmembrane proteins as well. We have upgraded our Dense Alignment Surface (DAS) method, so the new DAS-TMfilter method is capable to identify helical transmembrane proteins from large datasets with high speed and accuracy and provides the transmembrane segments of these integral membrane proteins.

Essentially, DAS-TMfilter performs the protein selection in the following way. The (true and/or false) transmembrane segments of a query sequence are used in our original DAS method to recognize transmembrane regions in a sequence library of transmembrane proteins of known topology. By testing the performance of the query sequence in "predicting" transmembrane segments of protein with known topology, the transmembrane or non-transmembrane character of the query protein can be identified rapidly with high accuracy. The analyses were also tested in several whole genomes too.