Identifying time-delayed gene regulatory networks via an evolvable hierarchical recurrent neural network

Table 8 The effect of noise level on GRN reconstruction in case of non-linearity between the genes. Results are the average of accuracy of the Link criteria for GRN reconstruction of 10 different randomly generated synthetic networks

Methods	σ ²	TP	FP	FN	Precision	Recall	F-score	Nominal p-value	Adjusted p-value
HRNN	0.1	20.2	28.3	19.0	0.41	0.51	0.46
	0.25	19.9	26.7	20.6	0.42	0.49	0.45
	0.5	17.9	28.4	21.3	0.38	0.45	0.41
	1	11.2	30.5	28.4	0.26	0.28	0.27
	1.5	5.9	32.2	33.6	0.15	0.14	0.15
TD-ARACNE	0.1	13.5	102.5	25.7	0.12	0.34	0.17	1.9×10⁻¹⁰	9.5×10⁻¹⁰
	0.25	14.7	107.3	25.8	0.11	0.36	0.16	1.4×10⁻⁸	7.0×10⁻⁸
	0.5	12.0	102.0	27.2	0.09	0.30	0.13	1.3×10⁻⁷	6.5×10⁻⁷
	1	14.5	129.5	25.1	0.10	0.37	0.15	3.0×10⁻⁴	1.5×10⁻³
	1.5	12.2	115.8	27.3	0.11	0.31	0.14	0.86	1
HCC-CLINDE	0.1	19.0	9.5	20.2	0.66	0.48	0.56	0.04	0.2
	0.25	18.0	10.3	22.5	0.63	0.44	0.52	0.08	0.4
	0.5	12.8	8.4	26.4	0.61	0.33	0.42	0.78	1
	1	4.1	9.4	35.5	0.29	0.10	0.15	0.01	0.05
	1.5	0.5	7.3	39.0	0.06	0.01	0.02	1.2×10⁻⁶	6.0×10⁻⁶
TSNI	0.1	14.1	139.9	25.1	0.09	0.36	0.14	6.2×10⁻¹¹	3.1×10⁻¹⁰
	0.25	13.6	128.4	26.9	0.09	0.33	0.14	7.2×10⁻¹⁰	3.6×10⁻⁹
	0.5	15.3	138.7	23.9	0.09	0.38	0.15	4.2×10⁻⁸	2.1×10⁻⁷
	1	17.6	160.4	22.0	0.09	0.44	0.16	4.8×10⁻⁴	2.4×10⁻³
	1.5	15.1	150.9	24.4	0.08	0.37	0.14	0.63	1
ebdbNet	0.1	7.8	56.2	31.4	0.12	0.20	0.14	1.5×10⁻⁹	7.5×10⁻⁹
	0.25	7.7	58.3	32.8	0.11	0.18	0.13	8.9×10⁻¹⁰	4.4×10⁻⁹
	0.5	5.2	48.8	34.0	0.09	0.13	0.10	4.4×10⁻⁹	2.2×10⁻⁸
	1	6.2	55.8	33.4	0.10	0.15	0.12	5.0×10⁻⁵	2.5×10⁻⁴
	1.5	5.2	54.8	34.3	0.07	0.13	0.09	0.02	0.1

The p-values are for how the Link F-scores of other methods compare with HRNN. σ ² is variance of the noise. Networks include 20 genes

ISSN: 1756-0381