Skip to main content

Table 2 The LPbyCD parameter optimization via internal cross-validation on benchmark data sets and how it compares to the setting when not discovering communities at all

From: LPbyCD: a new scalable and interpretable approach for Link Prediction via Community Detection in bipartite networks

Data set

CD method/measure

Internal validation

External validation

AUC

\(\sigma\)

AUPR

\(\sigma\)

AUC

\(\sigma\)

AUPR

\(\sigma\)

Enzyme

w/o CDa/JC\(_{NC}\)

0.88

0.01

0.11

0.01

Spectral part./JC\(_{CC}\)

0.85

0.00

0.14

0.04

0.85

0.01

0.19

0.07

Spectral part./JC\(_{NC}\)

0.91

0.00

0.19

0.02

0.92

0.01

0.26

0.06

Louvain/JC\(_{CC}\)

0.94

0.00

0.52

0.00

0.96

0.00

0.69

0.04

Louvain/JC\(_{NC}\)

0.95

0.00

0.56

0.02

0.96

0.01

0.71

0.02

GPCR

w/o CDa/JC\(_{NC}\)

0.78

0.02

0.18

0.02

Spectral part./JC\(_{CC}\)

0.79

0.01

0.11

0.02

0.80

0.02

0.15

0.02

Spectral part./JC\(_{NC}\)

0.84

0.01

0.19

0.01

0.85

0.01

0.25

0.02

Louvain/JC\(_{CC}\)

0.73

0.01

0.18

0.02

0.80

0.03

0.31

0.05

Louvain/JC\(_{NC}\)

0.85

0.01

0.28

0.03

0.89

0.01

0.49

0.04

IC

w/o CDa/JC\(_{NC}\)

0.85

0.01

0.25

0.02

Spectral part./JC\(_{CC}\)

0.82

0.01

0.31

0.03

0.83

0.02

0.40

0.04

Spectral part./JC\(_{NC}\)

0.87

0.00

0.32

0.06

0.88

0.01

0.41

0.05

Louvain/JC\(_{CC}\)

0.94

0.00

0.50

0.01

0.95

0.01

0.65

0.04

Louvain/JC\(_{NC}\)

0.96

0.00

0.61

0.02

0.97

0.01

0.78

0.03

NR

w/o CDa/JC\(_{NC}\)

0.68

0.08

0.16

0.05

Spectral part./JC\(_{CC}\)

0.73

0.02

0.21

0.03

0.72

0.06

0.24

0.08

Spectral part./JC\(_{NC}\)

0.75

0.03

0.19

0.02

0.77

0.06

0.23

0.13

Louvain/JC\(_{CC}\)

0.79

0.03

0.27

0.06

0.82

0.06

0.45

0.13

Louvain/JC\(_{NC}\)

0.81

0.03

0.30

0.05

0.80

0.09

0.42

0.14

Kinase

w/o CDa/JC\(_{NC}\)

0.85

0.01

0.32

0.03

Spectral part./JC\(_{CC}\)

0.76

0.01

0.17

0.01

0.77

0.03

0.23

0.02

Spectral part./JC\(_{NC}\)

0.85

0.00

0.26

0.02

0.86

0.01

0.35

0.02

Louvain/JC\(_{CC}\)

0.72

0.00

0.15

0.01

0.73

0.01

0.19

0.01

Louvain/JC\(_{NC}\)

0.90

0.00

0.38

0.01

0.91

0.01

0.50

0.03

  1. The results for spectral partitioning and the Louvain algorithm used as a community detection (CD) method. The results between internal and external validation align very closely w.r.t. AUC (highlighted in bold)
  2. aAssuming m = 0 for spectral partitioning as CD method