Skip to main content

Table 3 Results of testing our model, DeepBind and catRAPID on RNA sequences of 51 nucleotides

From: Predicting protein-binding regions in RNA using nucleotide profiles and compositions

 

#RBP-binding

      

RBP

RNA regions

Sensitivity

Specificity

Accuracy

PPV

NPV

MCC

Our model

       

FUS

100

79.00%

70.00%

74.50%

72.48%

76.92%

0.492

FXR1

97

88.66%

70.00%

79.19%

74.14%

86.42%

0.596

FXR2

93

69.89%

70.00%

69.95%

68.42%

71.43%

0.399

IGF2BP2

94

55.32%

70.00%

62.89%

63.41%

62.50%

0.256

LIN28A

96

58.33%

70.00%

64.29%

65.12%

63.64%

0.285

QKI

100

78.00%

70.00%

74.00%

72.22%

76.09%

0.482

TARDBP

100

22.00%

70.00%

46.00%

42.31%

47.30%

–0.091

Weighted average

 

64.41%

70.00%

67.25%

67.59%

66.94%

0.345

DeepBind

       

FUS

100

32.00%

33.00%

32.50%

32.32%

32.67%

–0.350

FXR1

97

32.99%

42.00%

37.56%

35.56%

39.25%

–0.251

FXR2

93

43.01%

73.00%

58.55%

59.70%

57.94%

0.168

IGF2BP2

94

48.94%

59.00%

54.12%

52.87%

55.14%

0.080

LIN28A

96

36.46%

53.00%

44.90%

42.68%

46.49%

-0.107

QKI

100

82.00%

81.00%

81.50%

81.19%

81.82%

0.630

TARDBP

100

50.00%

86.00%

68.00%

78.12%

63.24%

0.386

Weighted average

 

46.62%

61.00%

53.91%

53.73%

54.05%

0.077

catRAPID

       
  

DP value

     

FUS

10

16.40%

–

–

–

–

–

FXR1

10

17.60%

–

–

–

–

–

FXR2

10

22.30%

–

–

–

–

–

IGF2BP2

10

16.70%

–

–

–

–

–

LIN28A

10

19.10%

–

–

–

–

–

QKI

10

15.50%

–

–

–

–

–

TARDBP

10

18.10%

–

–

–

–

–

Weighted average

 

18.22%

–

–

–

–

–

  1. Sensitivity is shown for our model and DeepBind, and discriminative power (DP) value is shown for catRAPID. The specificity of our method is the same for all RBPs because it used a same set of negative data for all RBPs with a single model, whereas DeepBind has distinct models for each RBP. Due to the speed of the catRAPID server, catRAPID was tested on 10 RBP-binding sequences of 51 nucleotides for each RBP, whereas both our model and DeepBind were tested on all the RBP-binding sequences. Detailed results are available in Additional file 12