Simulation study.

<p><b>(A)</b> Comparison on static synthetic networks. From top to bottom, lines correspond Precision-Recall curves of four different methods. <i>Dashed black</i>: Hierarchical model trained by MCMC sampling. <i>Solid black</i>: Hierarchical model trained by variational approximation. <i>Solid blue</i>: Hypergeometric method <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0008118#pone.0008118-Goldberg1" target="_blank">[17]</a>. <i>Solid red</i>: MCODE <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0008118#pone.0008118-Bader2" target="_blank">[16]</a>. <b>(B)</b> Comparison on dynamic synthetic networks. From top to bottom, lines denote correspond to F scores over time frames. <i>Blue circle</i>: DYHM with . <i>Black squre</i>: DYHM with . <i>Green triangle</i>: DYHM with . <i>Red diamond</i>: DYHM with . <i>Dashed green</i>: Hypergeometric method <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0008118#pone.0008118-Goldberg1" target="_blank">[17]</a> applied separately to each each time frame.</p>