系统流行病学中网络差异比较的统计分析方法研究

Traditional epidemiology is usually referred to as “black box" epidemiology. To open the ‘black box‘ and illustrate the disease related biological pathways/networks, systems epidemiology based on high-throughput-omics technologies has created providing new insights. a The comparison of biological networks between groups (e.g. case vs. control, intervention vs. control) in population level is considered as the core of systems epidemiology, aming at identifying the disease related biological pathways and their effect on contributing to disease. However, effective methodology of network comparison is still lacking in systems epidemiology. In this project, targeting at the feature of the nodes, edges and direction changed simultaneously in the networks, an innovative and adaptive strategy of “Structural dissection → Statistical integration” is proposed. We will construct 3 statistics for detecting the undirected network/pathway difference, with “nodes + edges”, “edges ×weighted nodes” and “integrated score of nodes and edges”. Furthermore, we will construct a statistic integrating the nodes, edges and direction to detect the difference of the directed network/pathway that in principle capture the nodes and edges change, while accounting for the topology structure through putting on more weights on the difference of nodes locating on relatively more important role. Then, statistical simulations will be conducted to examine the stability and power of these 4 statistics, and acquire estimates of parameters by theoretical proof. Network data of cardiovascular disease and esophageal carcinoma will be analyzed to evaluate the practicability of the above proposed statistics. We hope our statistical methodologies of detecting networks difference in systems epidemiology will be constructed through this project, and it will help infer disease pathway, design drug targets, predict prognosis of diseases, as well as develop and evaluate intervention strategies.

传统流行病学被称为“黑盒子”流行病学。基于高通量组学技术的系统流行病学为打开黑盒子阐明危险因子致病通路/网络提供了机遇，在人群中比较组间(疾病VS.健康、干预VS.对照等)网络差异以推断危险因子致病通路及效应是其核心所在。然而，目前尚缺乏系统流行病学网络比较方法。本项目针对网络节点、边、方向共变特征，提出“结构拆分→数理整合”创新策略：采用节点+边、边×加权节点、节点与边整合得分策略构建3类无向网络/通路差异比较统计量；通过定义网络内上下游节点间调控权重，构建融节点、边和方向差异为一体的有向网络/有向通路比较统计量。利用统计模拟评价其稳定性、检验效能和优劣性，理论证明获得其参数统计量，实证分析心血管病、食管癌的系统流行病学网络数据验证其实用性。以期构建系统流行病学网络差异比较的统计方法体系及其R软件包，为推断致病通路，设计药物靶点，预测疾病发生、预后，制定及评估干预策略提供新方法。

探讨疾病危险因素进而推断病因是流行病学研究的永恒主题。系统流行病学方法已被广泛应用于人群流行病学研究中。其核心是是在人群水平上通过 “暴露因子-组学生物标记-疾病终点”网络/通路的组间差异统计学比较，推断危险因子导致疾病发生、发展或转归的网络或通路及其效应大小。本项目在系统流行病学理论框架内，针对人群研究中生物网络间的差异囊括 “节点”、“边”和“方向”的共变特征，形成“系统流行病学网络差异比较的统计分析方法体系”。主要完成以下5项工作：1）构建两个队列数据库。依托山东省20多家慢性病健康管理中心构建山东多中心健康管理纵向观察队列。构建山东食管癌病例-对照研究队列。2）完成无向网络(及特定通路)差异比较的检验统计量的构建：基于“节点信息+边信息”、“边信息×加权节点信息”和“节点信息与边信息整合得分”3种融合策略，采用在 “结构拆分→数理整合”基本思想，基于得分检验的统计量用于检验加权生物网络的组间差异，构建统计量NetDifM等多种检验方法。3）完成有向网络(特定有向通路)差异比较的检验统计量的构建：构建 “节点平均水平的差异”、“节点间连接状态的差异”和“边的方向性差异”融为一体的网络差异组间比较的WNES等统计量。利用预测因子与疾病表型之间的网络结构信息，提出疾病筛检模型判别能力的创新建模策略，为充分利用网络信息提高疾病筛检模型判别能力提供了新思路。4）探索了在系统流行病学中当复杂网络结构和潜在因果关系未知时，各种偏倚调整策略对暴露对结局因果效应估计的影响，同时提出了一系列偏倚调整的准则。并完成了编制涵盖上述统计量的各种具体检验方法的 “系统流行病学网络差异比较的R软件包”，供系统流行病学研究者应用。5）成果推广应用：应用所构建的统计量完成“美国Bogalusa心血管病队列研究的系统流行病学分析”和“中国山东食管癌病例-对照研究的系统流行病学分析”，以验证其实际应用价值。