High-order/variable-order dynamic Bayesian networks and dynamic qualitative probabilistic networks --- new models of gene regulatory networks

网络系统生物学(NSB)是一个新兴的领域，主要研究各种生物分子网络，如基因调控网络(GRNS)、转录调控网络(TRNS)、药物基因调控网络(DGNs)、蛋白质相互作用网络(PPI和DDIS)、代谢网络和信号通路网络。NSB研究的目标是通过分析和挖掘不同的分子相互作用，在高通量技术产生的联网数据的基础上，获得生物学知识，了解细胞系统和过程的机制。例如，从基因表达时间序列数据重建GRN或分析大型生物分子网络需要复杂的计算方法。由于生物分子的高度多样性及其相互作用的复杂性，高通量数据集的大尺寸和高维度往往包含测量中的不精确度、不确定性或噪声，或者样本或时间点较少，使得NSB的研究任务变得困难。 高阶和变阶动态贝叶斯网络(HO-DBN和VO-DBN)和动态定性概率网络(DQPN)是最近出现的GRN模型，它们能够预测不同基因之间的时延相互作用。DQPN是动态贝叶斯网络的定性抽象，也是定性概率网络的时间扩展，描述了复杂动态系统中的变量如何相互影响，以及对数据中的不精确性和不确定性具有很强的鲁棒性。这项研究的长期目标是：(1)设计高效的高阶DBN(HO-DBN)、变阶DBN(VO-DBN)和DQPN方法，用于从(多个)时间序列微阵列数据中识别准确的GRN，这些方法是可扩展的，能够处理数据中的噪声、不确定性和不精确度，并利用各种生物学知识；(2)设计高效的HO-/VO-DBN和DQPN方法，用于使用时间序列微阵列数据对给定的GRN进行修正、建模和模拟。

Network Systems Biology (NSB) is an emerging area focusing on the study of various types of biomolecular networks, such as Gene Regulatory Networks (GRNs), Transcriptional Regulatory Networks (TRNs), Drug Gene Regulatory Networks (DGNs), Protein Interaction Networks (PPIs and DDIs), Metabolic Networks, and Signaling Pathways Networks. The goal of NSB research is to gain biological knowledge and obtain an understanding of the mechanisms of cellular systems and processes, through analyzing and mining the different molecular interactions, and on the basis of the networked data generated from high-throughput technologies. Sophisticated computational methods are required to reconstruct GRNs from gene expression time-series data or to analyze large biomolecular networks, for instance. The task of NSB research is made difficult by the highly diverse set of biomolecules and the complexity of their interactions, the large size and high-dimensionality of high-throughput data sets often containing imprecision, uncertainty or noise in measurements, or small number of samples or time-points. High-Order and Variable-Order Dynamic Bayesian Networks (HO- and VO-DBNs) and Dynamic Qualitative Probabilistic Networks (DQPNs) are recent models of GRNs, which are capable to predict the time-delayed interactions between different genes. DQPNs are qualitative abstractions of Dynamic Bayesian Networks as well as temporal extensions of Qualitative Probabilistic Networks, describing how variables in a complex dynamic system influence each other and which are very robust to the imprecision and uncertainty in the data. The long-term goals of this research proposal are to: (1) Devise efficient High-Order DBN (HO-DBN), Variable-Order DBN (VO-DBN), and DQPN approaches for identifying accurate GRNs from (multiple) time-series microarray data, which are scalable and capable of dealing with noise, uncertainty and imprecision in the data, and using various biological knowledge; and (2) Devise efficient HO-/VO-DBN and DQPN methods for the revision, modeling, and simulation of given GRNs with time-series microarray data.