Different SMILES inputs result in different scores

[resulelgin]

Hello there!

Thank you very much for the tool you provided - we use it with great joy. During our trial runs we realized that the scores of docking can be different depending on the SMILES code we use for the same molecule even though everything else stays the same. We failed to understand what sort of SMILES input is recommended (e.g., Pubchem EOCHEM 2.3.0, RDkit, etc.). Do you have a suggestion for us? Any input is highly appreciated!

Ex:
Bezafibrate Methyl Ester SMILES from Pubchem: CC(C)(C(=O)OC)OC1=CC=C(C=C1)CCNC(=O)C2=CC=C(C=C2)Cl -> Aggregate score: 0.62
Bezafibrate SMILES from RdKit: COC(=O)C(C)(C)Oc1ccc(CCNC(=O)c2ccc(Cl)cc2)cc1-> Aggregate score: 0.56

[arogozhnikov]

Hi Resul,

we use it with great joy

glad to head that!

Regarding your question - that's an interesting observation, we trained on whatever used in PDB, which apparently is a mixture of SMILEs formats.

If you can run a dozen of complexes and check if there is a clear winner for smiles format, that would be super helpful for other users too.

[resulelgin]

Hi Alex,

Thanks for your answer. Unfortunate that we did not conventionalize databases like PDB for these purposes yet. I will keep you updated about my benchmark!