NEUROGEN - Neuronale und genetische Algorithmen auf Parallelrechnern in der Genomforschung. Teilvorhaben: Erkennung von orts- und feldbezogenen Eigenschaftsmustern in Proteinsequenzen Abschlussbericht

We developed a method based on hierarchical self-organizing maps (SOMs) to recognize patterns in protein sequences. The method is fully automatic, does not require prealigned sequences, is insensitive to redundancy in the training set, and works surprisingly well even with small learning sets. As it uses unsupervised neural networks, it is able to extract patterns that are not present in all of the unaligned sequences of the learning set. The identification of these patterns in sequence databases is sensitive and efficient. The procedure has been successfully applied to a number of notoriously difficult cases with distinct recognition problems: helix-turn-helix motif in DNA-binding protein, the CUB domain of developmentally regulated proteins and the superfamily of ribokinases. A comparison with established database search procedures such as PROFILE led to the conclusion that the new automatic method performs as well as or better than existing procedures. We consider the development of a knowledge-based visualization and analysis tool as necessary for analyzing the rapidly increasing amount of information in a comprehensive manner. We have shown that the current, and future, protein data collection can be characterized on the basis of multiple, local motifs. The protein sequences, containing over 40,000 characteristic conserved regions, were stored in an associative network. The memory is self-organizing and associative. The protein sequences were then projected with a non-linear transformation onto a 2-D map, whereby homologies were expressed as neighbor relationships. This process provides a simultaneous classification of the sequence regions, and, through a surface area projection, a clearer depiction of local and global sequence relationships. This new type of comparative analysis offers an efficient method for rapid interpretation of the data, particularly with regard to the genome sequencing programs for several multicellular organisms. Functionally related sequences, as well as functionally diverse proteins, can be elucidated using the principle of multiple motifs and modular protein architecture. With the appropriate WWW-Browser, the user receivers a quick data analysis of either nucleotide or peptide sequences within a few seconds (http://www.mdc-berlin.de/home.html target=NewWindow> http://www.mdc-berlin.de/home.html ). The interactive network is on-line and was implemented on a massive parallel computer (MP-1216 with 16385 prozessoren); an update process can be done in less than one hour. (orig.)Available from TIB Hannover: F97B2074+a / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekSIGLEBundesministerium fuer Bildung, Wissenschaft, Forschung und Technologie, Bonn (Germany)DEGerman