ABOUT

Dataset:
There were some dataset used in EnPPIpred. Such as : i) Postive dataset, ii) Negative dataset I, iii) Negative dataset II, iv) Blind postive dataset & v) Blind negative dataset. The positive E. coli TAP-tag pull-down experimental protein-protein interactions (PPIs) data were downloaded from Bacteriome.org (Hu et al. TAP interaction dataset (AKA 'Core - experimental')). The negative E. coli protein pairs were generated by random pairing between proteins.

Methods:
Support Vector Machine (SVM) with 5-fold cross-validation technique was employed, to build a model for predicting protein-protein interactions (PPIs) in Enteropathogen. Different features including domain-domain associations (DDA), degree (No. of interacting partners present in positive protein-protein interactions network), amino acid composition (AAC) and dipeptide composition (DC) were tested in order to optimize the performance of proposed SVM model.

Table 1: SVM kernel-wise performance measures of E.coli dataset using 5-fold cross-validation technique. Optimal parameters(DDA, degree and amino acid composition (Default Hybrid)) and threshold were used for respective kernel.

 

SVM Kernel	Sensitivity (%)	Specificity (%)	Accuracy (%)	PPV (%)	MCC	F1 Score (%)	ROC
Linear	73	89	81	87	0.63	79.39	0.881
Polynomial	49	96	72	92	0.51	63.94	0.769
Radial basis function	77	86	82	85	0.64	80.80	0.878
Sigmoid	80	82	81	81	0.62	80.50	0.877

Figure 1: The ROC curves showing threshold independent performances of different kernels of SVM on E.coli dataset.

 

 

Go to top

 

Table 2: SVM performance measures on different subsets of features in E. coli dataset, using RBF kernel. Optimal parameters were used for respec-tive subset of features.

 

Features of Protein Pair Features	Vector Length	Sensitivity (%)	Specificity (%)	Accuracy (%)	PPV (%)	MCC	F1 Score (%)	Area under ROCcurve
Domain-Domain Association(DDA)	1	32	92	62	80	0.57	45.71	0.633
Degree	1	83	79	81	80	0.62	81.47	0.874
DDA and Degree	2	80	82	81	81	0.62	80.50	0.874
Amino acid composition (AAC)	40	78	76	77	77	0.55	77.50	0.824
DDA and AAC	41	70	87	78	84	0.57	76.36	0.844
Degree and AAC	41	77	85	81	84	0.63	80.35	0.878
DDA, Degree and AAC	42	77	86	82	85	0.64	80.80	0.878
Dipeptide Composition (DC)	800	79	76	77	77	0.55	77.99	0.826
Degree and DC	801	83	79	81	80	0.62	81.47	0.868
DDA, Degree and DC	802	76	86	81	85	0.63	80.25	0.868
DDA, AAC and DC	841	71	88	79	85	0.60	77.37	0.849
All Features (DDA, Degree,AAC and DC)	842	76	87	82	85	0.63	80.25	0.882

 

Go to top

 

Table 3: Different threshold wise SVM performance measures of E.coli dataset using 5-fold cross-validation, parameters t = 2 (RBF kernel), g = 0.001, c = 0.9, j = 1 and best subset of features (DDA, Degree and AAC)

 

Threshold	Sensitivity(%)	Specificity(%)	Accuracy(%)	PPV(%)	MCC	F1 Score(%)
2.00	0	20	10	20	0.01	0.00
1.90	0	40	20	30	0.01	0.00
1.80	0	80	40	73	0.02	0.00
1.70	0	80	40	63	0.02	0.00
1.60	1	100	50	84	0.04	1.98
1.50	2	100	51	90	0.08	3.91
1.40	3	100	51	91	0.11	5.81
1.30	6	99	53	91	0.16	11.26
1.20	13	99	56	93	0.24	22.81
1.10	28	98	63	93	0.36	43.04
1.00	47	97	72	93	0.50	62.44
0.90	57	95	76	92	0.57	70.39
0.80	62	94	78	91	0.59	73.75
0.70	66	93	79	90	0.61	76.15
0.60	69	91	80	89	0.62	77.73
0.50	71	90	81	88	0.62	78.59
0.40	73	89	81	87	0.63	79.39
0.30	74	88	81	87	0.63	79.98
0.20	75	88	81	86	0.63	80.12
0.10	76	87	82	85	0.64	80.25
0.00	77	86	82	85	0.64	80.80
-0.10	78	85	81	84	0.63	80.89
-0.20	79	84	81	83	0.63	80.95
-0.30	80	83	81	82	0.62	80.99
-0.40	81	82	81	82	0.63	81.50
-0.50	82	81	81	81	0.62	81.50
-0.60	83	79	81	80	0.62	81.47
-0.70	84	78	81	79	0.62	81.42
-0.80	86	75	81	78	0.61	81.80
-0.90	88	72	80	76	0.60	81.56
-1.00	91	61	76	70	0.55	79.13
-1.01	92	58	75	69	0.53	78.86
-1.02	92	55	74	67	0.51	77.53
-1.03	92	51	72	66	0.48	76.86
-1.04	93	48	71	64	0.46	75.82
-1.05	94	44	69	63	0.43	75.44
-1.06	95	39	67	61	0.40	74.29
-1.07	95	34	65	59	0.37	72.79
-1.08	96	30	63	58	0.34	72.31
-1.09	97	26	61	56	0.32	71.01
-1.10	97	21	59	55	0.28	70.20
-1.20	99	3	51	51	0.09	67.32
-1.30	100	1	50	50	0.06	66.67
-1.40	100	0	50	50	0.03	66.67
-1.50	100	0	50	50	0.01	66.67
-1.60	100	0	50	50	0.00	66.67
-1.70	100	0	50	50	0.00	66.67
-1.80	100	0	50	50	0.00	66.67
-1.90	100	0	50	50	0.00	66.67
-2.00	100	0	50	50	0.00	66.67

 

Go to top

 

Table 4: Performance measures on blind E. coli protein-protein interactions (PPIs) dataset (337 positive set was obtained from the Protein Data Bank (PDB) and 337 negative set was obtained from random pairs), using proposed optimal SVM model.

Threshold	Sensitivity(%)	Specificity(%)	Accuracy(%)	PPV(%)	MCC	F1 Score(%)
1.00	6	99	47	92	0.14	11.27
0.90	9	99	49	89	0.17	16.35
0.80	11	98	49	89	0.18	19.58
0.70	13	98	50	87	0.19	22.62
0.60	14	97	51	86	0.19	24.08
0.50	15	97	51	85	0.20	25.50
0.40	17	96	52	85	0.21	28.33
0.30	19	96	53	86	0.22	31.12
0.20	19	96	53	85	0.22	31.06
0.10	19	96	53	85	0.22	31.06
0.00	20	95	53	85	0.23	32.38
-0.10	21	95	53	84	0.23	33.60
-0.20	22	95	54	85	0.24	34.95
-0.30	24	95	55	85	0.25	37.43
-0.40	24	94	55	84	0.25	37.33
-0.50	27	94	56	85	0.27	40.98
-0.60	28	93	57	84	0.27	42.00
-0.70	31	93	58	85	0.30	45.43
-0.80	33	93	60	86	0.32	47.70
-0.90	36	92	61	84	0.32	50.40
-1.00	47	85	63	79	0.33	58.94
-1.01	49	83	64	79	0.34	60.48
-1.02	51	82	65	78	0.34	61.67
-1.03	53	78	64	76	0.32	62.45
-1.04	55	73	63	72	0.28	62.36
-1.05	58	68	62	70	0.26	63.44
-1.06	61	64	62	68	0.25	64.31
-1.07	64	59	62	66	0.23	64.98
-1.08	66	49	59	62	0.15	63.94
-1.09	71	42	58	61	0.14	65.62
-1.10	75	38	58	60	0.13	66.67
-1.20	96	4	56	56	0.01	70.74
-1.30	100	0	56	56	0.00	71.79
-1.40	100	0	56	56	0.00	71.79
-1.50	100	0	56	56	0.00	71.79

Fig 2: Frequency plot of average GO semantic similarity of HPS and Random S. Typhi PPIs.

Go to top

 

 

Citations: Barman RK, Jana T, Das S, Saha S Prediction of Intra-species Protein-Protein Interactions in Enteropathogens Facilitating Systems Biology Study PLoS One. 2015 Dec 30;10(12):e0145648   PMID: 26717407