b

graphical structure Integrative and systems biology is a very promising tool for deciphering the biological and ge-netic mechanisms underlying complex traits. In particular, gene networks are used to model interactions between genes of interest. They can be defined in various ways, but a standard approach is to infer a co-expression network from genes expression measured by means of se-quencing techniques (for example, microarrays). Among methods used to perform the inference, Gaussian graphical models (GGM) are based on the assumption that the gene expressions are distributed as Gaussian variables, and Σ is their covariance matrix. Non-zero partial cor-relations between two genes are modeled by network edges, and are directly obtained from the inverse of Σ. But it turns out that estimating the inverse of Σ leads to an ill posed problem, since this kind of data leads to a number of observations (typically less than one hundred) that is usually much smaller than the number of variables (the number of genes/nodes in the network can range from a few hundred to several thousands). To overcome this difficulty, the seminal papers [8, 9] were the basis for the package GeneNet, in which the partial correlation is estimated either by means of a bootstrap approach (not available in the package anymore