jwohlwend
The problem of strange ligand conformations predicted by Boltz-2 through SMILES format.
The conformations of my ligand pridict by boltz-2 is strange，What could be the reason?
this is the SMILES ：CC(=NC#N)N(C)CC1=CN=C(C=C1)Cl
The name of the compound is acetamiprid.
Thank you for your reply! I used these two parameters: --use_potentials and --affinity_mw_correction. If I remove both of them, the predicted conformation looks like this, which seems even more unreasonable.
Did this happen when using or not using the --use_potentials flag?使用或不使用 --use_potentials 标志时是否发生这种情况？
https://github.com/jwohlwend/boltz/issues/532#issuecomment-3166722384
When testing the same yaml file, I'm not seeing the same issue. Do you have this issue with other complexes as well? It might also be worth trying to delete the preprocessed files.
在测试相同的 yaml 文件时，我没有看到相同的问题。您在其他综合体上也有这个问题吗？可能还值得尝试删除预处理的文件。
Thank you for your reply. Could you please tell me how to delete the preprocessed files? I don't quite understand. By the way, did you deploy boltz-2 on Windows? Because I downloaded it directly to my computer, and I wonder if that's the reason.
When running boltz predict 1.yaml you should see the creation of a new folder boltz_results_1 with subdirectories processed predictions, etc. When preprocessing your schema for structure prediction, these processed files get stored in boltz_results_1/processed and I would recommend deleting this directory or the entire boltz_results_1 directory and starting again.
Since MSA's are only used for protein sequences there is no need to run the command with the --use_msa_server flag and also while this isn't the source of your error, we currently only support binding affinity prediction for Protein - Ligand complexes so you won't be able to use Boltz2 Affinity for your DNA - Ligand complex.
If you see this issue across multiple different complexes (including the examples in the repo), I would suggest reinstalling boltz in a clean conda environment and trying again.
Thank you all for your answers.
The problem has been solved. I re-downloaded boltz-2 on WSL (Windows Subsystem for Linux, i.e., the Linux subsystem of Windows) and successfully predicted the reasonable conformation of the ligand. The reason is probably that I operated directly under the Windows system before.