Network Reconstruction Methods For Modeling Of Biological Systems

Reconstructing networks becomes one of the most fundamental aspects of SystemsBiology, as the innovation and application of high-throughput techniques generating moreand more measurement data for biological systems. In data-driven reverse-engineering ofgenetic regulatory networks and protein signaling pathways, the particular complexity ofthe biological data promotes further development of many mathematical methodologies,suchasdifferentialequations, probabilisticnetworks, soft-computingtechniques, Grangercausality, network statistics, etc. In this paper, we use fuzzy logic, Granger causality andlocal network statistic to tackle the problems of network reconstructing from the equilib-rium data, time series data, and protein sequence and structure data, respectively.To model the uncertainty of the equilibrium measurements in biological systems,soft-computing as a recently proposed uncertainty modeling technology has been em-ployed in reconstructing the protein signaling pathway, and the theoretic foundations aswell as the model solving methods of soft-computing techniques used in this paper havealso been investigated. Observed the conditional dependency and time varying volatilityin the time series data from biological systems, the well-known causality inference sta-tistical tool, Granger causality, has been modified and extended to heteroscedastic cases.Considering proteins as amino acids contacting networks, we have profiled the local net-work characteristics of protein-protein contact sites on protein surfaces, and combinedthe information from protein sequences and their 3-D structures to predict the protein-protein contact sites. The predictions on CheY in the chemotaxis pathway of Rhodobac-ter sphaeroides have been found matching the experimental results from the latest reportspublished in PLoS Biology. The paper is organized as followsIn the first chapter, the research background, the available biological data, and cur-rent research status of data-driven network reconstruction in Systems Biology are brieflyreviewed. The structure of this paper is also presented in this chapter.The second chapter firstly estimates the uniform approximation rates of a class offuzzy systems with kernel-shaped if-part fuzzy sets, and discusses the relationships be-tween the shapes of the if-part fuzzy sets and the approximation capability for the fuzzysystems. Next, we propose a co-evolution framework for particle swarm optimization, and demonstrate its robustness against the variation of the dimensions of some bench-mark functions. Based on the theoretic developments of the soft-computing techniques, anovel probabilistic fuzzy logic model is established for the inference of protein signalingpathway from equilibrium measurements in cells.In the third chapter, Granger causality has been modified to address the problem ofconditional dependency in the modeling of gene networks from time series expressiondata. Furthermore, observed the time varying volatility in the physiological data, we ex-tend Granger causality to detect causality in heteroscedastic data. In the view of causalityinference, it is the first time that the models of Granger causality in mean and Grangercausality in variance have been unified. The proposed model has been tested against asuit of toy models and also applied to the physiological recordings of Parkinson patients.Based on the multiple sequence alignments and 3-D structures of a pair of proteins, asuite of methods to predict the protein-protein contact sites has been developed by consid-ering proteins as networks with nodes of residues and edges of contacts between residuesin the fourth chapter. The methods have been used to predict the protein-protein contactsites between the proteins of chemotaxis pathway for Rhodobacter sphaeroides, and thesepredictions have been found to be consistent with the results in the latest experimentalreport published in PLoS Biology.Finally, some conclusions have been drawn and the further directions of this workhave also been listed in the last chapter.