Skip to main content

Table 2 Input parameters

From: MPRAP: An accessibility predictor for a-helical transmem-brane proteins that performs well inside and outside the membrane

Parameters Specificity Sensitivity Accuracy MCC
Single inputs
AA 0.57 0.89 0.59 0.19
R4S 0.70 0.71 0.68 0.37
Zpred 0.60 0.70 0.60 0.19
Zcoord 0.56 0.75 0.56 0.12
PSI scr 0.72 0.70 0.69 0.39
Combined inputs
R4S + Zpred 0.70 0.77 0.70 0.40
AA + Zpred 0.63 0.70 0.62 0.24
AA + R4S 0.70 0.78 0.71 0.41
AA + R4S + Zpred 0.71 0.77 0.71 0.43
PSI scr + Zpred 0.72 0.70 0.70 0.40
PSI scr + R4S 0.73 0.75 0.72 0.43
PSI scr + R4S + Zcoord 0.73 0.73 0.71 0.43
PSI scr + R4S + Zpred 0.74 0.74 0.73 0.44
Different kernels
PSI scr + R4S + Zpred (Linear) 0.72 0.74 0.71 0.41
PSI scr + R4S + Zpred (Polynomial) 0.73 0.74 0.72 0.44
PSI scr + R4S + Zpred (Radial basis) 0.74 0.74 0.73 0.44
Different probe radii
PSI scr + R4S + Zpred (1.4 Å) 0.74 0.74 0.73 0.44
PSI scr + R4S + Zpred (2.0 Å) 0.78 0.82 0.74 0.45
PSI scr + R4S + Zpred (2.0 Å/1.4 Å) 0.76 0.78 0.74 0.46
  1. Different combinations of input parameters used to train a Support Vector Machine to predict surface accessibility in a two state alphabet. For each predictor the specificity, sensitivity, accuracy and the Matthew Correlation Coefficient for predicting buried residues in a binary alphabet is reported. The first five lines contain the prediction results using a single type of information, where AA is amino acid encoded using sparse encoding, R4S is the substitution rate calculated from rate4site scores, Zpred is predicted distance from membrane center, Zcoord is the real (not predicted) distance from the membrane center and PSI scr is PSIBLAST-PSSM. The next group of predictors was obtained using combinations of these inputs. The next two lines contain the results for two predictors using the optimal combination of inputs but other kernels than the radial-basis kernel. The last line is the performance of the final version of MPRAP, i.e. the one trained to predict absolute accessibility.