BIOINFORMATICS<-->STRUCTURE
Jerusalem, Israel, November 17-21, 1996

Abstract

Artificial neural networks (ANN) for sequence and structure analysis of proteins

P.Wrede (1), G.Schneider (1), U.Hahn (2) and J. Schuchhardt (1)

(1) Freie Universität Berlin, Universitätsklinikum Benjamin Franklin, Institut für Med/Tech. Physik und Lasermedizin, AG Molekulare Bioinformatik Krahmerstr. 6-10, D-12207 Berlin, Germany
(2) Universität Leipzig, Institut für Biochemie, Talstr. 33, D-04103 Leipzig Germany

wrede@zedat.fu-berlin.de

Two differnt ANNs have been applied for the analysis of proteins: a) A supervised feedforward network for the identification of the cleavage site of signal peptides. The trained ANN was used as a fitness function for the simulated molecular evolution generating new sequences proven to be biologically avtive. b) Classification of local protein structural elements by a Kohonen network of 10X10 neurons. The network was trained to perform a topological mapping from the complex space of structural elements to the 2D plane.

Back to the Abstract Index.