Application Of Artificial Neural Network In Research In Bioinformatics

Bioinformatics is an emerging discipline which is generated by biology, mathematics, physics, chemistry, computer science, information science and so on.Today the main contents of bioinformatics are:analysis of genomic sequences, gene evolution, drug design, gene region prediction, gene function prediction and prediction of protein structure. Intelligent computing methods include artificial neural networks, hidden Markov models, support vector machines, genetic algorithms, ect. And they have unique advantages in the field of a large amount of data, noise model and lacking unified theory.The first chapter of this paper mainly introduces bioinformatics and the present research situation of some online learning methods. At the same time we analyse the advantages and disadvantages of various models, and point out the improvement direction. Also the main research work of this paper is summarized.Traditional model of gene regulatory networks often lack an effective method for solving gene expression data because of high time and space complexity. A new model for constructing gene regulatory networks using back propagation (BP) neural network was proposed in chapter 2. Combined with complex nonlinear mapping and self-learning, the BP neural network was mapped into a complex network. Through the statistics and comparison of network parameters based on urchin gene samples under different temperatures, diverse average degrees, average path lengths, modularities, average clustering coefficients, and map’s densities were observed. This indicated that certain gene expression data at different temperatures indeed changed. Differentially expressed urchin genes associated with temperature were found by calculating the difference in the degree of every gene from different networks.We carried out accurately similarity analysis to protein sequences in chapter 3. First of all, using Wang and Wang’s method the protein sequences were converted to 5- letter sequences, and the five letters were evenly distributed on the unit circle centered at the origin. Then we could obtain the position coordinates of the protein sequences. Sencondly combined with three physicochemical indexes of amino acids, an amino acid wase represented with a 5- dimensional vector. Finally, using self-organizing mapping neural networks for different protein vectors to make a clustering analysis, and it achieved better clustering effect.In the end of the thesis, we look forward to the future of data mining of gene expression profile and protein classification methods. The direction of future research is tentatively proposed according to the limitations and deficiency of the present method.