UniMol Docking v2 on PoseBusters quality checks

[liyue9129]

Hi !

I've tried to reproduce the results of UniMol Docking v2 on the PoseBusters dataset using the provided script, posebuster_demo.ipynb. The final RMSD values I obtained are similar to those reported in the literature.

`428it [00:03, 108.55it/s]

---

results length: 428
RMSD < 0.5 : 0.06542056074766354
RMSD < 1.0 : 0.2897196261682243
RMSD < 1.5 : 0.5934579439252337
RMSD < 2.0 : 0.7313084112149533
RMSD < 3.0 : 0.8294392523364486
RMSD < 5.0 : 0.9135514018691588
avg RMSD : 2.045954420285687

---

results length: 428
RMSD < 0.5 : 0.09345794392523364
RMSD < 1.0 : 0.40186915887850466
RMSD < 1.5 : 0.6495327102803738
RMSD < 2.0 : 0.7570093457943925
RMSD < 3.0 : 0.8434579439252337
RMSD < 5.0 : 0.9228971962616822
avg RMSD : 1.9020785139194096`

However, when I used bust to check the molecular quality, I found that the results differed significantly from those reported, with only 53.5% agreement. I'm wondering if I've made a mistake somewhere. There are 87 moleculars can't be read by RDKit with sanitize=True. Could you please provide the specific metrics you used for quality control or provide the code to check the molecular quality?

The metrics I used are as follows:

`

# chemical validity and consistency #
"mol_pred_loaded",
"mol_true_loaded",
"mol_cond_loaded",
"sanitization",
"molecular_formula",
"molecular_bonds",
"tetrahedral_chirality",
"double_bond_stereochemistry",

# intramolecular validity #
"bond_lengths",
"bond_angles",
"internal_steric_clash",
"aromatic_ring_flatness",
"double_bond_flatness",
"internal_energy",

# intermolecular validity #
"minimum_distance_to_protein",
"minimum_distance_to_organic_cofactors",
"minimum_distance_to_inorganic_cofactors",
"volume_overlap_with_protein",
"volume_overlap_with_organic_cofactors",
"volume_overlap_with_inorganic_cofactors",`

Thank you ！