Skip to main content

Advertisement

Table 2 Experimental comparison between SCPRED and competing structural class prediction methods.

From: SCPRED: Accurate prediction of protein structural class for sequences of twilight-zone similarity with predicting sequences

Test type Algorithm Feature vector (# features) Reference Accuracy MCC GC2
     all-α all-β α/β α+β overall all-α all-β α/β α+β  
Jackknife SVM (Gaussian kernel) CV (20) [36] 68.6 59.6 59.8 28.6 53.9 0.52 0.42 0.43 0.15 0.17
  LogicBoost with decision tree CV (20) [23] 56.9 51.5 45.4 30.2 46.0 0.41 0.32 0.32 0.06 0.10
  Bagging with random tree CV (20) [34] 58.7 47.0 35.5 24.7 41.8 0.33 0.26 0.22 0.06 0.06
  LogitBoost with decision stump CV (20)   62.8 52.6 50.0 32.4 49.4 0.49 0.35 0.34 0.11 0.13
  SVM (3rd order polyn. kernel) CV (20)   61.2 53.5 57.2 27.7 49.5 0.46 0.35 0.39 0.11 0.13
  Multinomial logistic regression custom dipeptides (16) [28] 56.2 44.5 41.3 18.8 40.2 0.23 0.20 0.31 0.06 0.05
  Information discrepancy1 dipeptides (400) [22, 24] 59.6 54.2 47.1 23.5 47.0 0.46 0.40 0.24 0.04 0.12
  Information discrepancy1 tripeptides (8000)   45.8 48.5 51.7 32.5 44.7 0.39 0.39 0.25 0.06 0.11
  Multinomial logistic regression custom (34) [27] 71.1 65.3 66.5 37.3 60.0 0.61 0.51 0.51 0.22 0.25
  SVM with RBF kernel custom (34)   69.7 62.1 67.1 39.3 59.5 0.60 0.50 0.53 0.21 0.25
  StackingC ensemble custom (34)   74.6 67.9 70.2 32.4 61.3 0.62 0.53 0.55 0.22 0.26
  Multinomial logistic regression custom (66) [26] 69.1 61.6 60.1 38.3 57.1 0.56 0.44 0.48 0.21 0.21
  SVM (1st order polyn. kernel) autocorrelation (30)   50.1 49.4 28.8 29.5 34.2 0.16 0.16 0.05 0.05 0.02
  SVM (1st order polyn. kernel) custom (58) [29] 77.4 66.4 61.3 45.4 62.7 0.65 0.54 0.55 0.27 0.28
  Linear logistic regression custom (58)   75.2 67.5 62.1 44.0 62.2 0.63 0.54 0.54 0.27 0.27
  SVM (Gaussian kernel) PSI-PRED based (13) this paper 92.6 79.8 74.9 69.0 79.3 0.87 0.79 0.68 0.55 0.55
  SVM (Gaussian kernel) custom (8 PSI-PRED based) this paper 92.6 80.6 73.4 68.5 79.1 0.87 0.79 0.67 0.54 0.54
  SCPRED custom (9) this paper 92.6 80.1 74.0 71.0 79.7 0.87 0.79 0.69 0.57 0.55
10-fold cross validation SVM (Gaussian kernel) CV (20) [36] 67.9 59.1 58.1 27.7 53.0 0.51 0.42 0.41 0.14 0.16
  LogicBoost with decision tree CV (20) [23] 51.9 53.7 46.5 32.4 46.1 0.38 0.37 0.31 0.07 0.10
  Bagging with random tree CV (20) [34] 53.5 51.0 37.6 22.0 41.2 0.28 0.30 0.22 0.04 0.06
  LogitBoost with decision stump CV (20)   63.2 53.5 50.9 32.4 50.0 0.48 0.36 0.36 0.12 0.14
  SVM (3rd order polyn. kernel) CV (20)   61.4 54.0 55.2 27.4 49.2 0.46 0.35 0.37 0.10 0.13
  Multinomial logistic regression custom dipeptides (16) [28] 56.9 44.2 42.2 17.7 40.2 0.24 0.20 0.32 0.04 0.06
  Multinomial logistic regression custom (34) [27] 69.9 65.3 66.5 38.4 60.0 0.60 0.52 0.51 0.23 0.25
  SVM with RBF kernel custom (34)   70.2 61.6 67.6 39.6 59.8 0.60 0.49 0.53 0.22 0.25
  StackingC ensemble custom (34)   73.4 67.3 69.1 29.8 59.9 0.59 0.52 0.54 0.18 0.25
  Multinomial logistic regression custom (66) [26] 69.1 60.5 59.5 38.1 56.7 0.56 0.44 0.48 0.20 0.21
  SVM (1st order polyn. kernel) autocorrelation (30)   52.4 49.7 0.3 30.4 35.1 0.18 0.16 0.05 0.06 0.02
  SVM (1st order polyn. kernel) custom (58) [29] 77.7 66.8 60.7 45.4 62.8 0.64 0.54 0.54 0.28 0.28
  Linear logistic regression custom (58)   74.7 66.4 62.7 45.8 62.4 0.63 0.54 0.54 0.27 0.28
  SVM (Gaussian kernel) PSI-PRED based (13) this paper 93.2 79.5 75.7 69.4 79.7 0.87 0.79 0.70 0.55 0.55
  SVM (Gaussian kernel) custom (8 PSI-PRED based) this paper 92.5 80.4 73.7 68.0 79.0 0.87 0.79 0.67 0.54 0.54
  SCPRED custom (9) this paper 92.8 80.6 74.3 71.4 80.1 0.87 0.79 0.70 0.57 0.56
  1. 1This method was not originally tested using 10-fold cross validation and thus we also did not report these results