Table 5.

Performances of the various GRN inference methods on the datasources. AUPR in the top 20 % of the possible connections with a undirected evaluation for each GRN inference method on the different datasources of the benchmark with a 20 % local Gaussian noise and 10 % of global lognormal noise. The best statistically significant results tested with a Wilcoxon test are highlighted for each datasource. Results obtained with current version (1.0) of the package and are updated online

Datasource		ARACNE	C3NET	CLR	GeneNet	Genie3	MRNET	MutRank	MRNETB	PCIT	Zscore	Random
R1	mean	0.004	0.002	0.005	0.140	0.024	0.005	0.042	0.005	0.177	0.140	<0.001
	σ(×10−3)	1.1	0.789	1.22	16	2.97	1.26	7.27	1.26	16.1	13.6	0.0265
S1	mean	0.039	0.032	0.139	0.062	0.134	0.109	0.063	0.118	0.060	0.028	0.001
	σ(×10−3)	8.02	7.92	1.98	8.25	3.51	9.45	2.25	5.83	1.44	13.8	0.211
S2	mean	0.006	0.006	0.042	0.013	0.036	0.021	0.021	0.021	0.01	0.003	<0.001
	σ(×10−3)	1.19	1.63	1.55	1.56	1	2.76	0.959	2.01	0.522	1.46	0.1
G1	mean	0.106	0.100	0.139	0.085	0.108	0.134	0.034	0.084	0.063	0.001	<0.001
	σ(×10−3)	7.46	7.58	7.83	2.91	6.66	9.48	2.26	3.27	2.69	0.15	0.0141
G2	mean	0.101	0.095	0.106	0.037	0.069	0.126	0.025	0.058	0.044	<0.001	<0.001
	σ(×10−3)	11.4	9.95	4.49	1.62	3.44	9.49	1.43	2.23	2.16	0.0917	0.0265

p < 0.05