基于概率模型检测的基因调控网络优化控制问题研究

发布时间：2018-06-25 19:52

本文选题：基因调控网络 + 优化控制　；参考：《南京航空航天大学》2017年硕士论文

【摘要】：基因调控网络是一类基本且重要的生物网络,通过对其控制可以实现生物系统功能的调节。在系统生物学中,通过外部的干预控制构造关于基因调控网络的控制理论成为一个非常热门的主题。该控制理论有助于发展未来的基因治疗技术以及解决一些生命科学问题。目前,布尔网络和概率布尔网络作为重要的网络模型已经被广泛地用于基因调控网络优化控制问题的研究中。基因调控网络的优化控制问题是指寻找有效的控制策略使得网络按照期望的方式演变,同时最小化控制成本。由于采取不同的控制时长,优化控制问题可以分为有限范围优化控制问题和无限范围优化控制问题。根据两类优化控制问题的定义,本文提出了以概率模型检测技术为基础的解决方法。针对有限范围的优化控制问题,本文提出了一种以概率模型检测为基础求解最优期望总成本以及相应优化控制策略的方法。首先,该方法使用概率模型检测器PRISM提供的建模语言描述在具有控制输入情况下带有干扰且上下文相关的概率布尔网络。然后,归约优化控制中定义的最优期望总成本为马尔科决策过程上的最小可达性回报。最后,在PRISM中,reward结构用于描述优化控制中的定量信息,最小可达性回报属性使用概率计算树逻辑公式表达并且通过模型检测算法实现自动求解。获取到的有限范围优化控制策略能够有效地改变网络的短期运行行为。提出的方法被应用到细胞凋亡网络和WNT5A网络中。实验结果证明了方法的正确性和有效性。进一步,本文还考虑了使用该方法求解带有硬性约束条件的有限范围优化控制问题,并在实验中证明了方法的灵活性。针对无限范围的优化控制问题,本文提出了一种以遗传算法和概率模型检测器PRISM相结合的方式求解最优期望总成本以及相应优化控制策略的方法。无限范围控制策略是一种固定控制策略,即不依赖控制时间,确定在每个网络状态上的控制决策。首先,该方法归约无限范围控制中定义的期望总成本为离散时间马尔科夫链上的平稳状态回报。其次,构建包含固定控制策略的带有干扰且上下文相关概率布尔网络的PRISM模型。然后,为了求解无限范围优化控制问题,固定控制策略被编码为遗传算法解空间中的一个个体。个体的适应度值在PRISM中被计算。最后,遗传算法在解空间上迭代地执行遗传操作从而获取最优解。获取到的无限范围优化控制策略能够有效地改变网络长期运行行为。提出的方法被应用到WNT5A网络中。实验结果证明了方法的正确性和有效性。
[Abstract]:Gene regulation network is a kind of basic and important biological network, which can regulate the function of biological system by controlling it. In systems biology, it is a hot topic to construct the control theory of gene regulation network through external intervention control. The control theory can help to develop future gene therapy technology and solve some life science problems. At present, Boolean networks and probabilistic Boolean networks, as important network models, have been widely used in the study of genetic regulation network optimization control problem. The optimal control problem of genetic control network is to find an effective control strategy to make the network evolve in the desired way while minimizing the control cost. Because of the different control time, the optimal control problem can be divided into finite optimal control problem and infinite optimal control problem. According to the definition of two kinds of optimal control problems, this paper presents a method based on probabilistic model detection technology. For the problem of finite range optimal control, this paper presents a method to solve the optimal expected total cost and corresponding optimal control strategy based on probabilistic model detection. Firstly, the probabilistic Boolean network with interference and context is described by using the modeling language provided by the probabilistic model detector (PRISM). Then, the optimal expected total cost defined in reductive optimization control is the minimum reachability return in the Markov decision process. Finally, in PRISM, the reward structure is used to describe the quantitative information in the optimal control. The minimum reachability return attribute is expressed by the probabilistic computational tree logic formula and solved automatically by the model checking algorithm. The obtained finite range optimal control strategy can effectively change the short-term operation behavior of the network. The proposed method has been applied to apoptotic networks and WNT5A networks. The experimental results show that the method is correct and effective. Furthermore, this paper also considers the use of this method to solve the finite range optimal control problem with rigid constraints, and proves the flexibility of the method in experiments. To solve the problem of infinite optimal control, this paper presents a method to solve the optimal expected total cost and corresponding optimal control strategy by combining genetic algorithm and probabilistic model detector (PRISM). Infinite range control strategy is a kind of fixed control strategy, which is independent of control time and determines the control decision in each network state. Firstly, the expected total cost defined in the reduced infinite range control is the stationary state return on the discrete time Markov chain. Secondly, a PRISM model with interference and context-dependent probabilistic Boolean networks with fixed control strategies is constructed. Then, in order to solve the infinite optimal control problem, the fixed control strategy is encoded as an individual in the solution space of genetic algorithm. The fitness of individuals is calculated in PRISM. Finally, the genetic algorithm iteratively performs genetic operations on the solution space to obtain the optimal solution. The obtained infinite range optimal control strategy can effectively change the long-term operation behavior of the network. The proposed method is applied to WNT5A network. The experimental results show that the method is correct and effective.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：Q811.4

【参考文献】