Skip to main content

Table 4 Link prediction performance on the BioKG dataset (in percent)

From: BioBLP: a modular framework for learning on multimodal biomedical knowledge graphs

Method

Pretrained

MRR

H@1

H@3

H@10

TransE

 

20.61

12.15

22.14

36.27

    + BioBLP-P

 

14.76

8.42

15.60

25.92

    + BioBLP-P

\(\checkmark\)

14.87

8.47

15.67

26.16

    + BioBLP-M

 

6.56

4.46

6.78

9.90

    + BioBLP-M

\(\checkmark\)

8.72

6.22

9.42

13.10

    + BioBLP-D

 

7.42

4.91

7.39

11.62

    + BioBLP-D

\(\checkmark\)

15.74

11.73

16.44

23.04

ComplEx

 

42.75

31.55

48.09

65.50

    + BioBLP-P

 

16.53

11.13

17.17

26.92

    + BioBLP-P

\(\checkmark\)

34.88

25.11

39.52

54.83

    + BioBLP-M

 

1.68

1.24

1.86

2.49

    + BioBLP-M

\(\checkmark\)

1.78

1.37

1.93

2.54

    + BioBLP-D

 

0.01

0.00

0.00

0.02

    + BioBLP-D

\(\checkmark\)

0.45

0.36

0.43

0.55

RotatE

 

55.20

44.46

61.95

74.76

    + BioBLP-P

 

45.29

35.60

51.33

62.89

    + BioBLP-P

\(\checkmark\)

47.30

36.56

54.59

66.10

    + BioBLP-M

 

10.40

7.02

10.40

16.30

    + BioBLP-M

\(\checkmark\)

14.34

11.14

14.79

19.78

    + BioBLP-D

 

11.60

8.79

12.43

16.67

    + BioBLP-D

\(\checkmark\)

49.68

40.62

55.30

66.26