Skip to main content

Table 2 A representative list of performances of neural networks with different combinations of features used as inputs

From: Protein structure based prediction of catalytic residues

Feature Matthew’s correlation (%)
C D A K All residues Conserved > 3.5
X     16.28 22.05
  X    15.51 26.68
   X   10.12 27.56
X   X   18.71 28.16
  X X   19.78 32.48
X X    20.56 27.67
X X X   20.68 28.95
   X X 10.95 25.48
  1. C, sequence conservation; D, distance to the GCM; A, residue type; K, closeness. A check symbol on a column indicates that the feature was included in training. The Matthew’s correlation coefficient, expressed as a percentage, was calculated on the validation set after convergence of the training procedure. Networks were trained either with “All residues” or only with residues that showed a conservation value of 3.5 or more. The calculation of the MCC was performed over the entire original set of residues, regardless of the effect of preselection on the overall counts.