Gene network inference using continuous time Bayesian networks: a comparative study and application to Th17 cell differentiation

Table 1 Performance comparison of CTBNs, DBNs and GC on simulated data for different network sizes

Method	NETs size	Mean precision	Mean recall	Mean F ₁	F ₁SEM
GC	10	0.46	0.68	0.54	6.40E-02
	20	0.40	0.70	0.49	4.33E-02
	50	0.24	0.82	0.37	3.23E-02
	100	0.16	0.82	0.27	2.13E-02
DBNs	10	0.90	0.29	0.41	6.90E-02
	20	0.55	0.42	0.47	3.66E-02
CTBNs	10	0.66	0.58	0.61	5.13E-02
	20	0.72	0.48	0.57	2.79E-02
	50	0.53	0.57	0.54	1.95E-02
	100	0.45	0.51	0.48	2.28E-02
Random	10	0.16	0.55	0.24	2.12E-02
	20	0.11	0.51	0.18	1.68E-02
	50	0.03	0.49	0.06	4.35E-03
	100	0.02	0.50	0.04	1.15E-03
Method	NETs size	Mean precision	Mean recall	Mean F ₁	F ₁ SEM
GC	10	0.42	0.75	0.52	4.18E-02
	20	0.28	0.81	0.41	2.32E-02
	50	0.22	0.78	0.34	1.58E-02
	100	0.14	0.80	0.23	5.24E-03
DBNs	10	0.62	0.53	0.56	3.40E-02
	20	0.60	0.57	0.58	4.31E-02
CTBNs	10	0.95	0.58	0.69	6.08E-02
	20	0.72	0.70	0.70	3.86E-02
	50	0.64	0.56	0.59	3.84E-02
	100	0.56	0.51	0.53	2.65E-02
Random	10	0.18	0.59	0.27	2.10E-02
	20	0.07	0.49	0.12	1.27E-02
	50	0.05	0.50	0.08	4.88E-03
	100	0.02	0.50	0.05	2.63E-03

Organism E.coli (top) and S. cerevisiae (bottom). Aggregate F ₁, precision and recall values are calculated as the arithmetic mean over the sets of 10 sampled network instances, the standard error of the F ₁ mean (SEM) is also shown. See also Figure 3.

ISSN: 1471-2105