Table 1 Study on the performances of MMFRLUnimodality

From: Multimodal fusion with relational learning for molecular property prediction

Data Set

BBBP

Bace

Sider

Clintox

HIV

MUV

Tox21

ToxCast

ESOL

FreeSolv

Lipo

SMILES

92.9 ± 1.5

90.9 ± 3.3

64.9 ± 0.3

78.2 ± 1.9

83.3±1.1

80.1 ± 2.5

85.7±1.2

70.5 ± 2.5

0.811 ± 0.109

1.623±  0.168

0.539±  0.017

NMRspectrum

91.0  ± 2.0

93.2±2.7

68.1±1.5

87.7±6.5

80.9 ±  1.7

80.9±5.0

85.1 ± 0.4

71.1±0.8

0.844 ± 0.123

2.417 ± 0.495

0.609 ± 0.031

Image

93.1  ±  2.4

92.9±1.8

65.3 ±  1.5

86.2 ± 6.5

82.3±  0.6

78.7 ± 1.7

86.0±1.0

71.0±1.6

0.761±0.068

1.648 ± 0.045

0.537±0.005

Fingerprint

92.9 ± 2.3

91.7  ± 3.6

65.6±0.7

87.5±6.0

81.2  ± 2.5

82.9±3.1

85.3 ± 1.3

70.0 ± 1.4

0.808±0.071

1.437±0.134

0.565 ± 0.017

NMRPeak

93.4±2.7

89.3 ± 1.7

62.8 ± 2.1

86.1 ± 5.4

82.1 ± 0.4

75.4 ± 5.2

84.9 ± 1.0

70.6 ±  0.8

0.924 ± 0.083

1.707 ± 0.126

0.587 ± 0.021

Average

92.8  ±  1.9

91.4 ± 2.7

65.3 ± 2.0

85.0 ± 5.7

81.8 ± 2.2

79.4 ± 4.0

85.4 ± 0.9

70.6 ± 1.3

0.830 ± 0.094

1.766 ± 0.394

0.586 ± 0.048

Max

93.4 ± 2.7

93.2 ± 2.7

68.1 ± 1.5

87.7 ± 6.5

83.3 ± 1.1

82.9 ±  3.1

86.0 ± 1.0

71.1 ± 0.8

0.761 ± 0.068

1.437 ± 0.134

0.537 ± 0.005

No Pre-training

91.9 ± 3.0

85.2 ± 0.6

57.0 ± 0.7

90.6 ± 0.6

77.1 ± 0.5

78.6 ± 1.4

75.9 ± 0.7

63.7 ± 0.2

1.050 ± 0.008

2.082 ± 0.082

0.683  ± 0.016

  1. The best results are denoted in bold, and the second-best are indicated with underlining among the five modalities. The first eight tasks are for classification under evaluation of ROC-AUC, while the last three are for regression with evaluation of RMSE.