Deep learning methods for protein torsion angle prediction

BMC Bioinformatics

Table 1 The Mean Absolute Error (MAE) of different feature combinations with the DBRM method

Number of features	Feature combination^a	phi	psi	avg^b
1	*PSSM*	23.28	35.12	29.2
	8-state secondary structure (8stateSS)	25.12	33.52	29.32
	Contacts_number_15_classes (CN15)	25.58	37.26	31.42
	Error_distribution_of_fragment_based_angles (fragsion)	24.24	40	32.12
	3-state secondary structure (3SS)	25.8	38.95	32.38
	Contacts_number_1_real_value (CN1)	26.92	44.71	35.82
	7 physicochemical properties (7PC)	27.27	52.18	39.73
	Solvent_accessibility (SA)	29.15	53.84	41.5
	Disorder	30.8	64.69	47.75
2	*PSSM_8stateSS*	22.18	30.73	26.46
	PSSM_CN15	22.41	33.14	27.78
	PSSM_Fragsion	22.19	34.29	28.24
	PSSM_7PC	22.42	35.75	29.09
	PSSM_DISORDER	22.96	35.23	29.1
	PSSM_SA	23.47	35.53	29.5
3	*PSSM_8stateSS_7PC*	21.48	30.36	25.92
	PSSM_8stateSS_Fragsion	21.63	30.72	26.18
	PSSM_8stateSS_CN15	21.99	30.12	26.06
	PSSM_SS8_Disorder	22.91	31.08	27
	PSSM_8stateSS_SA	23.09	31.41	27.25
4	*PSSM_8stateSS_7PC_CN15*	21.48	30.27	25.88
	PSSM_8stateSS_7PC_SA	21.88	30.89	26.39
	PSSM_8stateSS_7PC_Disorder	22.17	30.97	26.57
	PSSM_8stateSS_7PC_Fragsion	22.08	31.11	26.595
5	*PSSM_8stateSS_7PC_CN15_Disorder*	21.54	29.94	25.74
	PSSM_8stateSS_7PC_CN15_SA	21.93	30.39	26.16
	PSSM_8stateSS_7PC_CN15_Fragsion	21.81	30.83	26.32
6	*PSSM_8stateSS_7PC_CN15_Disorder_Fragsion*	21.11	30.33	25.72
6	PSSM_8stateSS_7PC_CN15_Disorder_SA	22.24	30.60	26.42
7	*PSSM_8stateSS_7PC_CN15_Disorder_Fragsion_SA*	21.36	29.83	25.6

^aFeatures combination: for example “PSSM_8stateSS” represent the combination of PSSM and 8-state secondary structure as input features. The bold font denotes the best combination selected for a specific number of features in terms of the average MAE of phi and psi angles
^bavg.: Average of phi and psi values for each features combination

ISSN: 1471-2105