Figure 1

Precision-recall curves for simulated networks. Precision-recall curves for different strategies for setting the significance threshold for the variational Bayes method as a function of the posterior probability and the randomized lasso with stability selection as a function of the empirical recovery probabilities for stability selection. Twenty replicate networks with 1000 variables, 300 observations, and an average of 1.47 edges per node were simulated (see Methods for further details). The network reconstruction methods compared are as follows, vb^α: variational Bayes method with posterior probabilities averaged in both directions of regression, vb^β: variational Bayes method with posterior probabilities not averaged, ${\ell }_1^{\alpha }$: randomized lasso with stability selection with the number of false positives bounded below 1 and recovery probabilities averaged in both directions of regression, ${\ell }_1^{\beta }$: same as ${\ell }_1^{\alpha }$, except without averaging, ${\ell }_1^{\gamma }$: randomized lasso with stability selection with the penalty parameter chosen such that the number of false positives are bounded below 1000 and recovery probabilities averaged in both directions of regression, and ${\ell }_1^{\delta }$: same as ${\ell }_1^{\gamma }$, except without averaging.