Coupling Literature And Microarray Data To Build Gene Regulatory Networks

Modern biomedical researches focus on regulatory mechanisms at the molecular level, which includes normal phenotype (such as differentiation, differences and so on) and abnormal phenotypes (such as cancer) of tissue function. Effective gene therapy means are being found to improve health and to improve disease treatment. However biological methods (such as targeted biological experiments) only can be used to identify an interactions between genes. It is difficult to identify interactions between multiple genes. In a recent multi-gene studies, computational science has been applied to process large-scale microarray data. Gene regulatory networks are usually large, variety, complicated and it is difficult to understand intuitively. Mathematical models are proposed to simulate the cellular gene activity.There are two flaws in previous research work. One is the lack of study on interactions between multiple genes and discovery of new genes involved in the expression of a particular gene function. The other is the difference between experimental results with different methods due to the limits of the mathematical model, algorithm and computer technology. In addition genetic data is huge, complex and dynamic. So a new method based on literature and microarray data is proposed in this thesis. The existing medical literature publications, experimental data of gene expression and the machine learning methods are used in the derivation of gene regulatory networks. Revealing the relationship between genes, gene regulatory networks related to specific function expression, can provide support for the treatment of diseases and health-related service.This article includes the following three aspects. First, existing medical literature resources of U.S. National Library of Medicine were chosen for the main resource. Using the online medical literature analysis and retrieval system (MEDLINE) and marking tools GENIA Tagger, defining gene entity relation rules according to the characteristics of gene names and the relationship between genes, extracting gene entity relationships, the initial seed network was created. Second, the related genes in the microarray data were classified according to the similarity.They were added to the appropriate seed network and the extended network was built. Finally, the interaction between genes in extended network was quantified. Genetic algorithm was used to solve the possible interactions matrix between genes. Removing the genes which have no interactions or weak interactions with others in the extended network, a streamlined gene regulatory network was built. The experimental results of each step having reached the appropriate evaluation criteria, demonstrated the effectiveness of this method. The usefulness of gene regulatory network obtained with this method can be further examined in biomedical experiments.