Computational Gene Prediction By Combining Two Gene Finding Programs

Now gene prediction is a critical research area in computational biology. This thesis introduces the research work on predicting genes in human DNA sequences. Currently many gene finders can predict genes with an agreeable accuracy. However, either the methods the gene finders apply have the limitations, or the applications of the gene finders are not easy for the biologists and the researchers in medicine to use. Aiming at the two disadvantages, this research work has presented two algorithms. Based on the two algorithms, a client-friendly platform can be developed to predict gene accurately.Since two basic approaches for predicting genes cannot reach the requirement of discovering more novel genes, the combination methods and cross-species comparative sequence analysis have been proposed. As more and more genomes of some organisms have been known, cross-species comparative sequence analysis method has been more and more helpful. The thesis summarized and classified the main algorithms applied in the method. Furthermore, two gene finders using this method have been studied. Their architectures and experiments have been reviewed separately and then their architectures, applications and performance have been compared overall. Afterwards, a study has been conducted to overlook and categorize the combination methods, which are widely used. Three gene finders have been analyzed and a comprehensive comparison of them has been given depending on their architectures, performance and applications. From these researches on cross-species comparative analysis method and the combination methods, it is indicated that the gene finders that use the combination methods can be enhanced by applying cross-species comparative sequence analysis. Moreover, the gene finders, which use comparative sequence analysis and the combination methods, have been selected for the next step of the research. Additionally, in biology, it has been proved that the genomes of Mus musculus and Canis familiaris were very important for predicting human genes. Therefore, the thesis presents two algorithms to predict genes by combining the two chosen gene finders. One uses the combination methods and another applies cross-species comparative analysis method in which the genomes of Mus musculus and Canis familiaris have been compared. The algorithms have been tested. By comparing the testing results with the program using the combination methods, it has been showed that to some extent the algorithms improved the performance of the gene finder using either comparative sequence analysis or combination methods. In addition, the two algorithms have their own advantages on predicting different genetic information.