Jia Bochao, Liang Faming
Eli Lilly and Company, Lilly Corporate Center, Indianapolis, IN 46225, USA.
Department of Statistics, Purdue University, West Lafayette, 47907, IN, USA.
Stat. 2020 Dec;9(1). doi: 10.1002/sta4.271. Epub 2020 Jan 21.
Graphical models have been used in many scientific fields for exploration of conditional independence relationships for a large set of random variables. Although a variety of methods have been proposed in the literature for estimating graphical models with different types of data, none of them is applicable for jointly estimating multiple mixed graphical models. To tackle this problem, we propose a joint mixed learning method. The proposed method is very flexible, which works for various mixed types of data, such as those mixed with Gaussian, multinomial, and Poisson, and also allows people to incorporate domain knowledge into network construction by restricting some links to be included in or excluded from the networks. As an application, the proposed method is applied to pan-cancer network analysis for six types of cancer with data from The Cancer Genome Atlas. To our knowledge, this is the first work for joint estimation of multiple mixed graphical models.
图形模型已在许多科学领域中用于探索大量随机变量的条件独立性关系。尽管文献中已提出了多种方法来估计具有不同类型数据的图形模型，但没有一种方法适用于联合估计多个混合图形模型。为了解决这个问题，我们提出了一种联合混合学习方法。所提出的方法非常灵活，适用于各种混合类型的数据，例如与高斯、多项式和泊松混合的数据，并且还允许人们通过限制某些链接包含在网络中或从网络中排除来将领域知识纳入网络构建。作为一个应用，所提出的方法应用于六种癌症的泛癌网络分析，数据来自癌症基因组图谱。据我们所知，这是第一项联合估计多个混合图形模型的工作。
