Protein Binding Site Prediction by Combining Hidden Markov Support Vector Machine and Profile-Based Propensities
Table 2
Performance of HM-SVM based method with and without order profile propensities.
Dataset
Method
Sp %
Sn %
F1 %
Acc %
MCC
AUC %
Heterocomplex I
HM-SVM 1a
44.9
56.0
49.8
68.3
0.274
69.5
HM-SVM 2b
52.4
73.5
61.2
73.8
0.436
81.4
Homocomplex I
HM-SVM 1
45.4
60.0
51.70
69.7
0.309
72.2
HM-SVM 2
54.5
74.6
62.9
76.3
0.474
83.6
Mix I
HM-SVM 1
45.5
58.0
51.0
69.4
0.297
71.2
HM-SVM 2
53.5
74.0
62.1
75.0
0.455
82.5
Heterocomplex II
HM-SVM 1
54.0
56.7
55.3
68.0
0.305
70.7
HM-SVM 2
60.8
71.7
65.8
74.0
0.454
81.2
Homocomplex II
HM-SVM 1
53.3
60.1
56.5
70.1
0.340
73.4
HM-SVM 2
61.1
73.8
66.8
76.4
0.493
83.7
Mix II
HM-SVM 1
53.6
58.6
56.0
69.3
0.326
72.4
HM-SVM 2
61.0
72.7
66.3
75.2
0.474
82.4
Results of HM-SVM 1 on the six data sets are obtained from [13]. HM-SVM 1 represents the HM-SVM predictor with the basic feature set using PSSM and ASA features; bHM-SVM 2 represents the HM-SVM predictor with the feature set using PSSM, ASA, and order profile propensity features.