Skip to main content

Table 1 Performance of the four models trained on crystal and docked poses and tested also on crystal and docked poses (schemes 1 and 2) on the PDBbind v2007 benchmark. Comparing the same models from the two first blocks (crystal:crystal and crystal:docked) shows that the pose generation error also introduces a small degradation in the test set performance. Making the same comparisons between the second and fourth blocks shows that a substantial part of this error has been corrected

From: Correcting the impact of docking pose generation error on binding affinity prediction

Model

Training

Test

RMSE

SD

Rp

Rs

1 (Vina)

Crystal

Crystal

2.41

1.99

0.554

0.608

2 (MLR::Vina)

Crystal

Crystal

1.88

1.85

0.630

0.680

3 (RF::Vina)

Crystal

Crystal

1.66

1.59

0.744

0.752

4 (RF::VinaElem)

Crystal

Crystal

1.52

1.42

0.803

0.799

1 (Vina)

Crystal

Docked

2.02

1.98

0.557

0.597

2 (MLR::Vina)

Crystal

Docked

1.90

1.87

0.622

0.670

3 (RF::Vina)

Crystal

Docked

1.76

1.72

0.693

0.710

4 (RF::VinaElem)

Crystal

Docked

1.60

1.52

0.772

0.771

2 (MLR::Vina)

Docked

Crystal

1.91

1.88

0.618

0.648

3 (RF::Vina)

Docked

Crystal

1.74

1.69

0.705

0.716

4 (RF::VinaElem)

Docked

Crystal

1.58

1.45

0.794

0.790

2 (MLR::Vina)

Docked

Docked

1.86

1.83

0.640

0.667

3 (RF::Vina)

Docked

Docked

1.69

1.63

0.730

0.730

4 (RF::VinaElem)

Docked

Docked

1.55

1.45

0.795

0.789