项目名称: 基于强化学习的前列腺癌蛋白质间相互作用网络的模型及方法研究
项目编号: No.61303108
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 朱斐
作者单位: 苏州大学
项目金额: 23万元
中文摘要: 前列腺癌是最主要的高发性恶性肿瘤之一,也始终是生物研究者的关注热点。分析前列腺癌蛋白质相互作用网络有助于加深对前列腺癌疾病机理、治疗和预防的理解。本项目研究如何从蛋白质图谱出发,使用强化学习的方法,建立一个前列腺癌的蛋白质相互作用网络。在该强化学习框架中,以每个阶段网络拓扑结构作为当前状态,以蛋白质作用的生物置信度作为奖赏,以选择交互作为动作。由于需要处理数百万个潜在节点的大规模网络,本项目建立一个适用于大规模空间问题的基于核方法的强化学习模型,并使用自适应归一化径向基函数作为势函数来塑造奖赏优化算法。最后,通过从生物学角度解析网络,提出描述前列腺癌相关网络的特征和数理模型。本项目通过蛋白质图谱建立网络可以有效扩充网络的规模,充分发挥强化学习方法的开放性以融合生物领域知识,利用强化学习在未知环境探索最优决策的特性以保障网络的稳定性和最优性。本项目的方法也为其他生物问题的求解提供了新的思路。
中文关键词: 强化学习;系统生物学;前列腺癌;蛋白质相互作用;复杂网络
英文摘要: Prostate cancer is one of malignant tumors with the highest incidence, and is always a hot spot of biologists. Analyzing prostate cancer protein interaction networks will help to deepen the understanding, treatment and prevention of prostate cancer disease mechanisms. This project studies how to construct prostate cancer protein interaction networks starting from protein map by using reinforcement learning. In the reinforcement learning framework, the network topology of each stage is regarded as the current state, bio-confidence of protein interaction is treated as a reward, and interaction choice is viewed as the action. As we have to to deal with large-scale networks which have millions of potential nodes, we establish a kernel-based reinforcement learning model which is suitable for large large-scale space problem,and use an adaptive normalized radial basis function as a potential function to shape the reward so as to optimize algorithm. Finally, through analyzing network from a biological point of view, we propose features and mathematical model which can be utilized to describe prostate cancer networks. The project effectively expands the sacle of the network through building network from protein map, integrate biological domain knowledge by giving full play to openness of reinforcement learning method, an
英文关键词: reinforcement learning;systems biology;prostate cancer;protein interaction;complex network