Learning Gaussian Graphical Models of Gene Networks with False Discovery Rate Control

In many cases what matters is not whether a false discovery is made or not but the expected proportion of false discoveries among all the discoveries made, i.e. the so-called false discovery rate (FDR). We present an algorithm aiming at controlling the FDR of edges when learning Gaussian graphical models (GGMs). The algorithm is particularly suitable when dealing with more nodes than samples, e.g. when learning GGMs of gene networks from gene expression data. We illustrate this on the Rosetta compendium [8].

• Type: Conference
• Year: 2008
• Where: EVOW
• Authors: Jose M. Peña
• Keywords: Artificial Intelligence, EVOW 2008, False Discovery, Gaussian Graphical Models, So-called False Discovery