Regulatory signal detection in genomic sequences.
Ho, Sy Loi.
Date of Issue2005
School of Computer Engineering
This thesis is to develop a general and robust approach for the detection of different types of regulatory signals in eukaryotic genomic DNA sequences. We proposed to use lower-order Markov models for encoding the input sequences for the prediction of signals by neural networks and demonstrated the efficacy of the Markov/neural approach in the detection of three signals, namely, splice sites, translation initiation sites, and transcription start sites. The low-order Markov models incorporate useful biological knowledge such as the differences in nucleotide distributions observed in different functional regions adjoining the signals and the homology of potential sites. The neural networks, being large non-parametric non-linear models, combine the outputs from Markov chains to derive long-range and complex interactions among nucleotides, which are related to the actual identification of signals. The novel Markov/neural hybrid mode was shown capable of approximating higher-order Markov models of signals, and provides an efficient and feasible method of learning model parameters, leading to the better detection of signals.
DRNTU::Engineering::Computer science and engineering::Computer applications::Life and medical sciences
Nanyang Technological University