Does CHAI1 Lose Ligand Chemical Bond Information in Predictions?

[binbinbinv]

Hi,
I encountered a potential issue with the CHAI1 tool regarding ligand structure predictions. Specifically, it seems that CHAI1 might be losing some chemical bond information during the prediction process.

Steps to Reproduce:
I used CHAI1 to predict protein-ligand complexes.
The input ligand file contained SMILES with a double bond (C=C) explicitly defined.
However, in the predicted output structure, the corresponding bond appeared as a single bond (C-C).

Could this be due to a limitation in CHAI1’s handling of ligand SMILES or the conversion process? Any insights or suggestions would be appreciated!

SMILES：

Output of ligand in complex structure：

[arogozhnikov]

However, in the predicted output structure, the corresponding bond appeared as a single bond (C-C).

model predict positions, not bonds between atoms. AFAIR we don't even write anything about bonds to CIF.