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

Table 9 The effect of noise level on GRN reconstruction in case of non-linearity between the genes

	Methods	σ ²	TP	FP	FN	Precision	Recall	F-score	Nominal p-value	Adjusted p-value
Delay	HRNN	0.1	18.7	34.2	20.5	0.35	0.47	0.41
		0.25	18.5	31.9	22.0	0.36	0.45	0.41
		0.5	15.6	34.4	23.6	0.31	0.39	0.34
		1	8.0	37.8	31.6	0.17	0.20	0.18
		1.5	4.5	37.3	35.0	0.11	0.11	0.11
Delay	HCC-CLINDE	0.1	18.3	10.2	20.9	0.64	0.46	0.53	0.01	0.05
		0.25	17.9	10.4	22.6	0.63	0.44	0.51	0.005	0.025
		0.5	12.7	8.5	26.5	0.60	0.32	0.42	0.03	0.15
		1	3.9	9.6	35.7	0.28	0.09	0.14	0.27	1
		1.5	0.4	7.4	39.1	0.05	0.01	0.01	3.9×10⁻⁶	1.9×10⁻⁵
Effect	HRNN	0.1	19.8	29.3	19.8	0.40	0.50	0.44
		0.25	19.2	27.6	21.9	0.41	0.46	0.43
		0.5	16.1	30.7	23.5	0.34	0.40	0.37
		1	9.8	32.3	29.9	0.23	0.24	0.23
		1.5	5.1	33.3	34.5	0.13	0.12	0.13
Effect	HCC-CLINDE	0.1	18.5	10.0	20.7	0.64	0.47	0.54	0.06	0.3
		0.25	18.0	10.3	22.5	0.63	0.44	0.52	0.02	0.1
		0.5	12.7	8.5	26.5	0.60	0.32	0.42	0.16	0.8
		1	3.9	9.6	35.7	0.28	0.09	0.14	0.02	0.1
		1.5	0.4	7.4	39.1	0.05	0.01	0.01	3.4×10⁻⁷	1.7×10⁻⁶

Results are the average of accuracy of the Delay and Effect criterion for GRN reconstruction of 10 different randomly generated synthetic networks. The p-values are for how the Delay and Effect F-scores of HCC-CLINDE method compare with HRNN. σ ² is variance of the noise. Networks include 20 genes

ISSN: 1756-0381