Neuronale und genetische Algorithmen auf Parallelrechnern in der Genomforschung. Teilvorhaben: Anwendung von parallelen Hoechstleistungsrechnern in der Genomforschung Abschlussbericht

The genetic information of organisms is conserved in DNA/RNA sequences. The one-dimensional DNA/RNA sequence can be seen as a string of letters in a four letter (nucleotide) alphabet. This letter code determines the three-dimensional structure and function of gene products. An important and difficult to solve question is what is the relation between one-dimensional sequence and three-dimensional structure/function. Biophysical/biochemical experiments are used to solve the problem. Newer approaches also use statistic and information processing methods. Neuronal networks are well suited as an interface between letter oriented information and structural properties. In our approach we apply hierarchical neural networks. An important aspect of our work is the use of parallel computers and algorithms for the neural network training of complex networks used in our approach. High performance algorithms and software for parallel computers, a special neurocomputer and workstation clusters have been developed. The neural network trianing speed of the Cray T3E system reaches 2.457 Billion Connection Updates Per Second (GCUPS) measured with standard NETtalk backpropagation benchmark. This is the highest performance for training neural networks worldwide (1997). A neural network system has been developed for splice site detection of human pre-mRNA sequences (100%/0.44% donor and 97.55%/1.17% acceptor correct/wrong positive detections). For the detection of eukaryotic promoter sequences heirachical multi-state neural networks with time delays (TDNN, MS-TDNN) have been applied with great success. For DNA sequence analysis the visualisation tool DIANA (Dna Interactive Artificial Neural network Analysis) has been developed. (orig.)SIGLEAvailable from TIB Hannover: F97B2339+a / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekBundesministerium fuer Bildung, Wissenschaft, Forschung und Technologie, Bonn (Germany)DEGerman