A Multiattribute Gaussian Graphical Model for Inferring Multiscale Regulatory Networks: An Application in Breast Cancer.
A Multiattribute Gaussian Graphical Model for Inferring Multiscale Regulatory Networks: An Application in Breast Cancer.
复制标题
用于推断多尺度调节网络的多属性高斯图形模型：在乳腺癌中的应用。
DOI：
--
复制
发表时间：
2018
影响因子：
--
通讯作者：
M. Sundqvist
中科院分区：
文献类型：
--
作者：
J. Chiquet;G. Rigaill;M. Sundqvist
本章通过整合多个数据来源来解决分子生物学中重建调控网络的问题。我们考虑从不同技术中测量的数据集，这些数据集都与同一组变量和个体相关。这种情况在分子生物学中变得越来越普遍，例如，当同一组患者的同一组“基因”相关的蛋白质组学和转录组学数据可用时。为了推断一个整合蛋白质组学和转录组学数据的共识网络，我们引入了高斯图形模型（GGM）的多元扩展，我们称之为多属性GGM。实际上，GGM框架为生物实体之间的直接联系建模提供了一个很好的代理。我们使用在多尺度水平上操作的邻域选择过程来执行多元GGM的推理。该过程采用组-拉索惩罚，以选择两个基因之间在蛋白质组学和转录组学水平上都起作用的相互作用。我们最终得到了一个共识网络，它嵌入了在细胞的多个尺度上共享的信息。我们用两个乳腺癌数据集来说明这种方法。r包可以在github上（https://github.com/jchiquet/multivarNetwork）公开获取，以提高可重复性。
This chapter addresses the problem of reconstructing regulatory networks in molecular biology by integrating multiple sources of data. We consider data sets measured from diverse technologies all related to the same set of variables and individuals. This situation is becoming more and more common in molecular biology, for instance, when both proteomic and transcriptomic data related to the same set of "genes" are available on a given cohort of patients.To infer a consensus network that integrates both proteomic and transcriptomic data, we introduce a multivariate extension of Gaussian graphical models (GGM), which we refer to as multiattribute GGM. Indeed, the GGM framework offers a good proxy for modeling direct links between biological entities. We perform the inference of our multivariate GGM with a neighborhood selection procedure that operates at a multiscale level. This procedure employs a group-Lasso penalty in order to select interactions which operate both at the proteomic and at the transcriptomic level between two genes. We end up with a consensus network embedding information shared at multiple scales of the cell. We illustrate this method on two breast cancer data sets. An R-package is publicly available on github at https://github.com/jchiquet/multivarNetwork to promote reproducibility.