Statistical analysis of microbiome longitudinal data with multiplex networks

网络通常是指一组离散实体及其交互的基于图形的表示。它们是分析生物系统及其组成部分之间复杂关系的有用概念工具;它们可以很容易地应用于微生物组数据，其中单个实体（节点）可以由个体，时间点或地理位置表示。将这些节点连接在一起的链接（边）可以表示遗传相似性或地理距离等。有几种方法可以研究微生物组纵向数据，无论是预测还是检测微生物群落的变化。与此同时，目前有许多网络方法可用于考虑时间序列日益复杂的情况。对于这个提议，我想通过（轴1）开发方法来有效地分析使用多路复用网络的微生物组纵向数据，然后（轴2）将这种网络技术应用于在不同时间尺度收集微生物组样本的各种经验项目中，来弥合这些领域之间的差距。也就是说，该提案分为五个不同的活动，在活动1.1中，我将使用模拟微生物组数据评估多路网络的性能，然后将这些方法用于实际的纵向数据。在活动1.2中，我将评估随机化模型，以分析作为时间图的多重网络，并将开发一个统计检验来实现这一点。在活动2.1中，我将分析为“1000次握手项目”收集的皮肤微生物组样本，以了解污染的动态。在活动2.2中，我将分析从孕妇身上收集的阴道微生物组样本，以评估社区的特定变化是否可以预测分娩。在活动2.3中，我将分析从捐赠给尸体农场的腐烂尸体上收集的微生物组样本，以估计死亡时间。如何从多层微生物组网络中提取有用的信息并评估其结构是研究人员感兴趣的问题。例如，网络中微生物之间的关系，而不是特定微生物个体的变异，可能是重要功能差异的基础。然而，我们缺乏统计框架来识别和充分比较不同微生物栖息地之间的网络。该提案旨在通过提出新的方法来满足这一需求，该方法使用多重时间网络来分析多个时间尺度上微生物群落的动态和稳定性。

Networks generally refer to graph-based representations of a set of discrete entities and their interactions. They are useful conceptual tools for analyzing biological systems and the complex relationships among their constituents; they can be easily applied to microbiome data in which individual entities (nodes) can be represented by individuals, time points, or geographic locations. Links (edges) connecting these nodes together can represent genetic similarities, or geographic distances, among others. There are several approaches to study microbiome longitudinal data, either to make forecasts or to detect shift in microbial communities. At the same time, there exist numerous network approaches currently available to take into account the increasing complexity of time series. For this proposal, I want to bridge the gap between such fields by (Axis 1) developing methods to efficiently analyze microbiome longitudinal data using multiplex networks, and then (Axis 2) applying such network techniques in a diversity of empirical projects for which microbiome samples have been collected at different time scales. Namely, the proposal is broken down in five distinct activities, In Activity 1.1, I will assess the performance of multiplex networks with simulated microbiome data, prior to using such methods to actual longitudinal data. In Activity 1.2, I will evaluate randomization models to analyze multiplex networks as temporal graphs and will develop a statistical test to do so. In Activity 2.1, I will analyze skin microbiome samples collected for the `1000 Handshakes project' to understand the dymamics of contamination. In Activity 2.2, I will analyze vaginal microbiome samples collected on pregnant women to assess whether specific changes in the communities can predict labor. In Activity 2.3, I will analyze microbiome samples collected on decomposing corpses donated to a body farm to estimate post-mortem interval. How to extract useful information from multilayer microbiome networks and assess their structure through time is a question of acute interest for researchers. For example, the relationships among microbes in a network, rather than variation in specific individual microbes, may underlie important functional differences. Yet, we lack the statistical framework to identify and fully compare networks among different microbial habitats over time. This proposal aims at filling this need by proposing novel approaches to analyze the dynamics and stability of microbial communities at multiple time scales using multiplex temporal networks.