Table 3. Prediction of regulatory types - E. coli data sets.

Method	Cold				Heat				Oxidative				Lactose diauxie
GGM	+	−	TN	TP	+	−	TN	TP	+	−	TN	TP	+	−	TN	TP
	0.027	0.016	0.994	0.006	0.029	0.019	0.994	0.002	0.029	0.017	0.995	0.004	0.026	0.022	0.995	0.004
ARACNE	—	—	0.937	0.088	—	—	0.963	0.102	—	—	0.969	0.063	—	—	0.887	0.242
CLR	—	—	0.968	0.028	—	—	0.968	0.039	—	—	0.968	0.033	—	—	0.968	0.027
L1	0.004	0	0.994	0.005	0.004	0	0.994	0.005	0.003	0	0.994	0.004	0.005	0	0.994	0.007
L1/2	0.003	0	0.994	0.005	0.003	0	0.994	0.005	0.003	0	0.994	0.004	0.007	0	0.994	0.006
L0	0.003	0	0.994	0.005	0.003	0	0.994	0.005	0.003	0	0.994	0.004	0.004	0	0.994	0.006
GENIE3	—	—	0.968	0.273	—	—	0.968	0.295	—	—	0.968	0.277	—	—	0.968	0.275
Global silencing	0.347	0.327	0.968	0.195	0.367	0.312	0.968	0.208	0.362	0.278	0.968	0.256	0.340	0.318	0.968	0.217
Network deconvolution	0.374	0.298	0.968	0.275	0.406	0.279	0.968	0.254	0.391	0.278	0.968	0.284	0.344	0.315	0.968	0.280
Proposed approach	0.247	0.156	0.968	0.246	0.247	0.156	0.968	0.256	0.247	0.156	0.968	0.264	0.247	0.156	0.968	0.257

The fraction of correctly predicted activating (+) and repressing (−) regulatory relationships with respect to the experimentally verified regulations from RegulonDB are presented for the four different data sets. The fractions of true positive (TP) and true negative (TN) edges irrespective of the regulatory type are also included. The ‘—’ is considered where the method is not able to infer the regulatory type