A Dynamic Model Of Gene Regulation With Reference To Multisource Biological Information

Gene regulation is a regulation upon gene expression involving various in-cell materials,and it is related to almost all cellular activities.Therefore,the study of gene regulation can not only explore the inherent laws of life activities,but also helps the prediction,diagnosis,treatment,and pharmacy of gene-related diseases.To facilitate such study,some network models were built by using multisource biological information(e.g.gene expression profiles,transcription factors,protein-protein interaction data and so on),and present the regulation among genes.However,the existing gene regulatory network models focus on only "gene-gene" relationships,rather than detailed regulatory mechanisms among various in-cell materials.And moreover,it also needs to develop the means,by which the multisource biological information is integrated into the models,so as to make full use of such information.Therefore,this paper presents a dynamic model of gene regulation with reference to multisource biological information.The main contributions of the thesis are described as follows.(1)A new and more precise dynamic model of gene regulationA model was built,focusing on the deep mechanism of gene regulation,in this paper.In such model,Petri nets were used to present the topology of the relationships among various in-cell materials.Compared with other models,the dynamic model is more precise and consistent with in-cell dynamics.(2)Using prior knowledges directlyBuilding a complex system model needs to integrate the prior knowledges from multisource biological information.However,the current gene regulatory models focus on "gene-gene" relationships only,the prior knowledges need to be transformed into some relationships between genes first,and then integrated indirectly.In this paper,the prior knowledges can be embedded in the model directly,so that they directly participate in solving the model,and the information within them could be fully explored.(3)Solving the model computationallySolving the model is to optimize a large number of parameters in a recursive network with numbers of hidden nodes,by using time series data and prior knowledges.In this paper,an objective function was defined,considering the integration of the prior knowledges,so as to avoid over-fitting.BP algorithm was then employed to solving the model.Moreover,in order to improve the performance of the algorithm,parallel technology was used.To verify the validity and universality of the dynamics model,an experiment was designed and conducted to test the ability of the model in adjusting and predicting gene expression behaviors.Meanwhile,the time consuming by using GPU parallel computing was compared with that by using CPU only.The results show that the parallel computing greatly improves the performance of the modeling process.