Skip to main content
. 2016 Jun 16;7:949. doi: 10.3389/fmicb.2016.00949

Table 3.

Performance (CV-10 fold) of SVM-based models on both balanced and realistic datasets using the composition-motif hybrid features.

Dataset Feature Kernel Thr Sen Spec Acc MCC AUC Parameter
Balanced dataset AAC_Motif hybrid t0 0.3 91.55 94.37 92.96 0.86 0.95 t:0,c:60
t1 0.2 90.14 97.18 93.66 0.88 0.95 t:1,d:1
  t2 0.9 90.14 97.18 93.66 0.88 0.97 g:0.001 c:0.05 j:4
DPC_Motif hybrid t0 0.4 87.32 95.07 91.2 0.83 0.97 t:0,c:990
t1 −0.4 93.66 97.18 95.42 0.91 0.96 t:1,d:2
    t2 0.1 92.25 95.77 94.01 0.88 0.97 g:0.001 c:1 j:1
Realistic dataset AAC_Motif hybrid t0 0.3 75.35 99.3 97.13 0.82 0.96 t:0,c:5
t1 −0.6 89.44 98.38 97.57 0.86 0.98 t:1,d:3
  t2 −0.3 88.73 98.67 97.77 0.87 0.98 g:0.001 c:4 j:1
DPC_Motif hybrid t0 0.5 78.87 97.89 96.17 0.77 0.95 t:0,c:990
t1 −0.3 81.69 99.02 97.45 0.84 0.97 t:1,d:2
t2 −0.3 85.92 99.02 97.83 0.87 0.97 g:0.001 c:2 j:1

The models constructed using balanced dataset displayed the highest MCC values of 0.88 and 0.91 for AAC_Motif hybrid and DPC_Motif hybrid models, respectively. The models constructed using realistic dataset displayed the highest MCC values of 0.87 for both AAC_Motif hybrid and DPC_Motif Hybrid models.