Collaborative Research: Bayesian Network Estimation across Multiple Sample Groups and Data Types

作为这项合作研究的一部分，研究人员将开发新的统计方法来估计多个图形网络。拟议的研究将解决当受试者和所考虑的变量之间存在异质性时学习网络的挑战，在图形建模和贝叶斯统计方面开辟新的天地。所开发的方法将有可能在统计和应用领域产生重大影响，在这些领域中自然会出现网络估计问题。特别是，将探索在神经成像中的应用。该项目将包括针对研究生的教育和培训活动。研究结果将传播给研究界，并用于进一步的跨学科合作努力。软件和代码将被开发并存放在公共存储库。当所有样本都是在类似条件下收集或反映单一类型的疾病时，可以应用图形套索或贝叶斯网络推理方法来学习潜在的条件依赖关系。然而，在许多研究中，不同亚型或疾病的样本是在不同的实验环境下或其他不同的条件下获得的。当考虑多种数据类型时，这一挑战变得更加艰巨。该项目将侧重于开发贝叶斯方法，以学习跨多个样本组的单一数据类型的网络，使用一种方法，即跨组链接边缘值，并灵活地对哪些组最相似。方法还将从不同的主题集和不同的数据类型扩展到网络的分层建模框架。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

As part of this collaborative research, the investigators will develop new statistical methods for the estimation of multiple graphical networks. The proposed research will address the challenge of learning networks when there is heterogeneity among both the subjects and the variables considered, breaking new ground in graphical modeling and Bayesian statistics. The methods developed will have the potential for significant impact in statistics and in applied fields in which problems of network estimation naturally arise. In particular, applications in neuroimaging will be explored. The project will include educational and training activities for graduate students. Findings will be disseminated to the research community and used to further interdisciplinary collaborative efforts. Software and code will be developed and deposited in public repositories.When all samples are collected under similar conditions or reflect a single type of disease, methods such as the graphical lasso or Bayesian network inference approaches can be applied to learn the underlying conditional dependence relations. In many studies, however, samples are obtained for different subtypes or disease, under varying experimental settings, or other heterogeneous conditions. The challenge becomes even more formidable when multiple data types are under consideration. This project will focus on the development of Bayesian methods to learn networks for a single data type across multiple sample groups using an approach that both links edge values across groups, and flexibly models which groups are most similar. Methods will also be extended to a hierarchical modeling framework of networks from both heterogeneous sets of subjects and heterogeneous data types.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.