Computational Promoter Recognition in Eukaryotic Genomic DNA

An important aspect within computational biology deals with the analysis of biological sequence data with methods known from statistical pattern recognition. This thesis describes a system to identify the regulatory DNA sequences known as promoters, which control the expression of genes in their proximity. Promoters follow a common structure because all the genes controlled by them are accessed by the same enzyme complex. However, individual promoters differ very much from each other: This enables a specific activation of a gene at a certain time or tissu, and thus the development of a complex organism. The thesis presents increasingly complex probabilistic models representing the DNA sequence and structure of promoters, and shows how they can be used to identify promoter regions in long DNA sequnces. Among other methods, different types of Markov chain and generalized hidden Markov models are studied, and a Bayesian classification approach is compared to neural networks. The system was successfully applied by the Drosophila Genome Project to the complete genome of the fruit fly, and results are compared with promoter recognition in human sequences.