Failed to generate 3D, replace with 2D

[selimsami]

Hi,

When I'm trying to do the preprocess_pka.py step, the protocol works well for a while and then it gets stuck at some molecule with "Failed to generate 3D, replace with 2D" error and keeps repeating that. Wonder if you experienced this and if you have any solutions?

Thanks,
Selim

python ./scripts/preprocess_pka.py --raw-csv-file Datasets/tsv/chembl_valid.tsv --processed-lmdb-dir chembl --task-name valid
raw_data size: 10000
12%|███████████████████▏ | 1188/10000 [01:39<21:52, 6.72it/s]Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D
Failed to generate 3D, replace with 2D

[ZhouGengmo]

Thanks for your interest in our work!

The issue is something we've experienced as well.

The "Failed to generate 3D, replace with 2D" warning occurs when RDKit is unable to generate a 3D conformation for certain molecules. In such cases, the pipeline automatically falls back to using 2D conformations instead. And each time this fallback happens, a warning is printed.

It seems like RDKit is struggling to generate 3D conformations for some specific molecules, which is causing the process to appear stuck. You can reduce the number of conformations or RDKit's max attempts parameter. This can help speed up the process for problematic molecules.

Or just wait it out. This might take some time depending on the complexity of the molecules.

[selimsami]

thanks! It does seem like the process is stuck at 1188/10000 with all these error messages, but what you're saying is that it's actually progressing even though the progress bar is perhaps stuck?

[ZhouGengmo]

Yes, some molecules may require tens of minutes or even more than an hour.