Skip to main content

Table 4 The RMSE (kcal/mol) of benchmark databases by DFT methods with respect to CCSD(T)/CBS benchmark interactions

From: A machine learning correction for DFT non-covalent interactions based on the S22, S66 and X40 benchmark databases

Methods

RMSE

S22 (DFT)

S22 (GRNN)

S66 (DFT)

S66 (GRNN)

X40 (DFT)

X40 (GRNN)

M062X/6-31G*(vac) (DFT1)

1.41

0.57

1.84

0.48

1.85

0.52

M062X/6-31G*a (DFT2)

2.55

0.58

1.67

0.48

1.31

0.52

M062X/6-31G*b (DFT3)

1.82

0.42

1.37

0.45

1.22

0.49

M062X/6-31+G*a (DFT4)

3.99

0.83

2.45

0.43

1.33

0.47

ωB97XD/6-31G*(vac) (DFT5)

1.68

0.30

1.46

0.43

1.71

0.53

ωB97XD/6-31G*a (DFT6)

2.36

0.60

1.70

0.50

1.25

0.55

ωB97XD/6-31G*b (DFT7)

1.56

0.28

1.46

0.43

1.35

0.56

B3LYP/6-31G*a (DFT8)

5.97

0.47

3.90

0.63

2.04

0.65

B3LYP-D3/6-31G*a (DFT9)

2.78

0.41

1.79

0.59

1.24

0.57

PBE/6-31G*a (DFT10)

4.66

0.57

3.07

0.67

1.80

0.56

PBE-D3/6-31G*a (DFT11)

2.68

0.45

1.79

0.61

1.66

0.60

  1. The best results are shown in italics
  2. vacThe calculations are performed in vacuum
  3. aThe solvent is set as water (ε = 78.35)
  4. bThe solvent is set as pentylamine (ε = 4.20), which possess a similar dielectric constant as the protein environment (ε ~4.0)
Back to article page

Journal of Cheminformatics

ISSN: 1758-2946

Contact us