Fig. 3.

LEM inference performance is robust to increasing network size and complexity. Oscillatory gene expression data were generated in silico for five-, ten-, and 20-node networks (a in silico 3; b in silico 4; c in silico 5, respectively). AUC-ROC scores (d, e, and f) were computed as described in Additional file 1: Section 5; recall that a random ranking of edges should produce an expected AUC-ROC score of 0.5. AUC area under the curve, LEM Local Edge Machine, ROC receiver-operating characteristic