Table 2 Comparative evaluation of SCANN, SCANN+, and six other DL models predicting five physical properties using the QM9 dataset.

From: Towards understanding structure–property relations in materials with interpretable deep learning

 

EHOMO

ELUMO

Egap

α

Cv

 

(meV)

(meV)

(meV)

(Bohr3)

(cal mol−1 K−1)

WaveScatt39

85

76

118

0.160

0.049

SchNet30

41

34

63

0.235

0.033

MEGNet35

38

31

61

0.081

0.030

Cormorant14

34

38

61

0.085

0.026

SE(3)-Trans15

35

33

53

0.142

0.054

ALIGNN36

21

19

38

0.056

–

SCANN

41

37

61

0.141

0.05

SCANN+

32

31

52

0.115

0.041

  1. The bold numbers denote the lowest mean absolute errors (MAEs) among the eight models.
  2. The dash symbol (–) indicates the result has not been reported yet.
  3. EHOMO Energy of the highest occupied molecular orbital, ELUMO Energy of the lowest unoccupied molecular orbital, Egap Energy gap, α Isotropic polarizability, Cv Heat capacity at 298 K.
Back to article page