Reverse Modling And Analysis Of Biological Network

How to accurately grasp information about relationships among genes from massive high-throughput data, construct appropriate formal models and study these relationships to capture the biology rules are important topics in post-genomic era. In this thesis, using genes' expression profiles data, the intrinsic mechanisms of lung cancer and Parkinson disease are discussed. The details are as follows:1. Using genes'expressions profiles data of lung tissue at normal and different types of cancer-adenocarcinoma (AC) and small cell lung cancer (SCLC), the corresponding gene mutual information networks are constructed. Through comparing the structural parameters of these networks, the significant difference of the structures of gene networks corresponding to normal and lung cancer is found. Further, according to the difference of degree, betweenness and core of each gene in these networks, nine structural key genes are obtained. Seven of them which are related to lung cancer are supported by literatures. We predict the remaining two genes AKT1 and REL may have important roles in the occurrence, development and metastasis of lung cancer. Then, gene logic networks of normal and different types of lung cancer are constructed. Comparing the distributions of 2-order logic types in these networks, we find that logic types in gene logic network of normal include logic types 1,3,5,8, while logic type 8 does not appear in that of both AC and SCLC, and logic type 5 does not appear in that of AC. Also, logic types 2,6 and 2,4,6 emerge in that of AC and SCLC respectively. According to the principle that a system's structure decides its functions, the difference of the distribution of 2-order logic types in the above gene logic networks of normal and different types of lung cancer may be a reason or result of the occurrence and development of lung cancer. It provides some enlightenment roles to study the intrinsic mechanism of lung cancer.2. The inherent law of living system can emerge after proper coarse graining. In the process of constructing biological network corresponding to the living system, choosing a proper threshold to coarse graining the network is an important premise to find intrinsic mechanism of organisms. We propose a new method to determine the threshold of gene network and apply it for studying the intrinsic mechanism of Parkinson disease. Using genes'expression data and phenotypes of Parkinson disease, phenotype networks of Parkinson disease are constructed from the view of gene and disease respectively according to the relationships between phenotypes and genes. Comparing the two types of phenotype networks, when their similarities reach to the maximum, the thresholds of gene network are 0.47 and 0.48. Furthermore, through calculating the specificity and susceptibility of phenotype networks, we observe the specificity and susceptibility also reach to the maximum at thresholds 0.47 and 0.48. The validity of these thresholds is verified. Besides, comparing the structural parameters of gene networks of normal and disease stage at different thresholds, significant difference between these two gene networks at threshold 0.47 or 0.48 is found. It further verifies the efficiency of this method. Finally, we study the common Gene Ontology terms of the related genes in the determined gene network. It shows the reliability of relationships among genes.3. According to the relationships between phenotypes and genes, analyzing the relationships and development of disease-related phenotypes is meaningful to the diagnosis and treatment of diseases. Based on the above determined gene network, logic network of Parkinson-related genes is constructed by logic analysis of phylogenetic profiles and then logic network of disease phenotypes is predicted according to the relationships between genes and phenotypes. 66 possible pathways are mined through analyzing disease phenotype logic network. Clinical data show that the development of motor symptoms in 49 of these pathways is consistent with clinical observations. Also, there are 33 pathways with non-motor symptoms as initial phenotypes which differ from clinical observations. The possible reason is that these non-motor symptoms are easy to be neglected when they are not serious. We predict that four non-motor symptoms-depression, anxiety, hallucination and personality changes are initial phenotypes of Parkinson disease. The method of mining development pathways of disease phenotype can also be applied the prediction of other disease. It provides helpful suggestions to diagnosis and treatment of disease in clinical studies.