Skip to main content

Table 2 Results

From: Prediction of DNA-binding proteins from relational features

   Accuracy    AUC  
PD138 vs. NB110 PF SF PSF PF SF PSF
Simple logistic regression 83.4 (1) 82.2 (2) 80.7 (3) 0.91 (2) 0.90 (3) 0.94 (1)
L2-regularized log. regression 81.4 (3) 83.5 (2) 85.5 (1) 0.92 (1) 0.91 (2) 0.91 (2)
SVM with radial basis kernel 81.8 (2) 79.9 (3) 85.1 (1) 0.92 (2) 0.90 (3) 0.93 (1)
Linear SVM 81.4 (3) 83.6 (2) 83.9 (1) 0.92 (2) 0.89 (3) 0.93 (1)
Ada-boost w. decision stamps 80.6 (2) 78.6 (3) 81.4 (1) 0.90 (1) 0.90 (1) 0.90 (1)
Random forest 81.8 (3) 83.5 (1) 82.3 (2) 0.90 (3) 0.91 (2) 0.93 (1)
Average ranking 2.33 2.17 1.5 1.83 2.33 1.17
UD54 vs. NB110 PF SF PSF PF SF PSF
Simple logistic regression 81.0 (3) 86.0 (1) 82.8 (2) 0.91 (1) 0.89 (2) 0.89 (2)
L2-regularized log. regression 82.2 (3) 82.4 (2) 84.1 (1) 0.89 (3) 0.91 (1) 0.90 (2)
SVM with radial basis kernel 81.0 (2) 84.0 (1) 80.4 (3) 0.92 (1) 0.88 (3) 0.91 (2)
Linear SVM 81.7 (2) 82.4 (1) 82.4 (1) 0.90 (2) 0.91 (1) 0.87 (3)
Ada-boost w. decision stamps 76.2 (3) 78.0 (2) 79.3 (1) 0.88 (3) 0.89 (2) 0.90 (1)
Random forest 78.6 (3) 79.3 (1) 79.2 (2) 0.88 (3) 0.89 (2) 0.90 (1)
Average ranking 2.67 1.34 1.67 2.17 1.67 2
BD54 vs. NB110 PF SF PSF PF SF PSF
Simple logistic regression 80 (3) 80.5 (2) 81.8 (1) 0.91 (1) 0.85 (2) 0.91 (1)
L2-regularized log. regres 83.1 (1) 81.9 (2) 81.7 (3) 0.92 (1) 0.88 (3) 0.91 (2)
SVM with radial basis kernel 82.5 (2) 82.5 (2) 83.6 (1) 0.91 (1) 0.90 (2) 0.90 (2)
Linear SVM 81.4 (3) 82.3 (2) 82.9 (1) 0.93 (2) 0.90 (3) 0.94 (1)
Ada-boost w. decision stamps 84.2 (1) 73.8 (3) 79.8 (2) 0.91 (1) 0.88 (2) 0.88 (2)
Random forest 82.4 (1) 75.0 (3) 79.4 (2) 0.89 (2) 0.89 (2) 0.91 (1)
Average ranking 1.83 2.33 1.67 1.33 2.33 1.5
APO104 vs. NB110 PF SF PSF PF SF PSF
Simple logistic regression 80.7 (3) 85.0 (1) 80.8 (2) 0.89 (3) 0.92 (1) 0.91 (2)
L2-regularized log. regression 82.6 (3) 84.5 (1) 83.1 (2) 0.90 (2) 0.91 (1) 0.91 (1)
SVM with radial basis kernel 79.4 (3) 83.2 (2) 84.1 (1) 0.88 (3) 0.90 (2) 0.91 (1)
Linear SVM 79.4 (3) 84.5 (1) 84.1 (2) 0.89 (2) 0.89 (2) 0.92 (1)
Ada-boost w. decision stamps 77.6 (3) 78.1 (2) 79.1 (1) 0.87 (2) 0.87 (2) 0.89 (1)
Random forest 81.7 (1) 78.5 (3) 79.4 (2) 0.88 (2) 0.87 (3) 0.89 (1)
Average ranking 2.67 1.67 1.67 2.33 1.83 1.17
ZF vs. NB110 PF SF PSF PF SF PSF
Simple logistic regression 95.1 (3) 98.7 (1) 97.2 (2) 0.99 (2) 1.0 (1) 1.0 (1)
L2-regularized log. regres 95.9 (3) 99.3 (2) 100 (1) 0.99 (2) 1.0 (1) 1.0 (1)
SVM with radial basis kernel 95.8 (3) 99.3 (1) 98.6 (2) 0.99 (2) 1.0 (1) 1.0 (1)
Linear SVM 81.4 (3) 99.3 (1) 97.8 (2) 1.0 (1) 1.0 (1) 1.0 (1)
Ada-boost w. decision stamps 95.9 (3) 99.3 (2) 100 (1) 0.98 (2) 1.0 (1) 1.0 (1)
Random forest 96.5 (3) 97.9 (1) 97.2 (2) 0.99 (2) 1.0 (1) 1.0 (1)
Average ranking 3 1.33 1.67 1.83 1 1
  1. Predictive accuracies and areas under the ROC curve (AUC) on 5 classification benchmarks achieved by 6 machine learning algorithms using physicochemical features (PF) as proposed by Szilágyi and Skolnick [8], structural features (SF) automatically constructed by our algorithm, and the combination of both feature sets (PSF).