Skip to main content

Table 7 The effect of data augmentation

From: Predicting protein-ligand binding residues with deep convolutional neural networks

 

Train

Aug-Train 1

TestSet 2

Precision

Recall

MCC

Precision

Recall

MCC

SITA

56.31

41.55

0.448

58.79

42.27

0.470

SIEX1

53.66

40.20

0.432

55.81

41.45

0.454

SIEX2

50.12

38.80

0.411

53.23

39.79

0.434

SIEX3

49.38

39.07

0.410

52.88

39.81

0.433

  1. 1Due to the difficulties in training on Aug-Train, networks with k=9 and N=10 are used in this experiment. The average-cross entropy loss per protein on Train and Aug-Train are 0.80 and 7.77, respectively. The cross-entropy loss on Aug-Train does not change substantially with further training. For fair comparison, we do not use more complex networks
  2. 2SIEX1: SITA-EX1, SIEX2: SITA-EX2, SIEX3: SITA-EX3