Skip to main content

Table 1 Overall prediction performance of the 15 RF classifiers and that of the ensemble with different votes on the CASP9 data set.

From: LigandRFs: random forest ensemble to identify ligand-binding residues from sequence information alone

Individual Ensemble
No. Sen(%) MCC Prec(%) F1(%) No. Sen(%) MCC Prec(%) F1(%)
1 57.15 0.30 25.25 29.83 1 87.86 0.19 9.98 16.85
2 59.40 0.35 31.13 33.59 2 84.48 0.22 12.58 19.95
3 63.24 0.31 24.35 29.38 3 81.86 0.24 14.05 21.49
4 65.76 0.33 25.20 31.14 4 80.28 0.25 15.39 22.67
5 44.50 0.32 34.65 31.24 5 78.84 0.27 16.65 24.09
6 57.83 0.31 26.22 30.47 6 75.99 0.27 17.38 24.58
7 59.12 0.33 29.19 31.62 7 74.81 0.28 18.86 26.00
8 59.23 0.32 27.18 31.16 8 73.32 0.29 20.21 27.03
9 67.88 0.30 22.80 22.56 9 72.63 0.30 21.17 28.04
10 51.21 0.31 28.62 31.21 10 71.21 0.31 23.42 29.34
11 46.99 0.31 30.96 30.53 11 69.51 0.32 24.69 30.19
12 64.50 0.30 23.61 28.96 12 67.31 0.33 25.72 30.64
13 61.25 0.31 25.28 29.20 13 64.93 0.33 26.68 31.07
14 40.51 0.31 38.38 30.41 14 62.01 0.34 28.78 32.46
15 59.98 0.31 26.08 30.50 15 56.96 0.37 34.97 35.99
  1. The left part shows the performance of each individual classifier, and the right shows the performance of the ensemble of the 15 classifiers with different votes, i.e., the ensemble predicts a residue to be ligand binding site if a number of RF classifiers predict it to be a ligand binding site residue. Here the ensemble with majority vote predicts a residue to be ligand binding site if all of the 15 RF classifiers predict it to be a binding residue.
  2. The italic number denotes the best performance by the measure of MCC.