tiejundong / flexpose Goto Github PK

Hi Authors,

Thanks for the brilliant work! I am a little confused with the training process, where apo structures, holo-structures, or fake apo structures are randomly selected as input. For flexible binding, is it more intuitive to input apo structures and predict holo-structures in the training process?

Thanks!

The web server only works when I use the provided input example but as soon as I use my protein and ligand it shows this error.

"FileNotFoundError: [Errno 2] No such file or directory: '/tmp/tmpdlk2psln/structure_output_path/0.pdb'
Traceback:
File "/data/lab_website/.conda/envs/streamlit/lib/python3.11/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 541, in _run_script
exec(code, module.dict)
File "/data/lab_website/lab_website_code/pages/1_FlexPose.py", line 141, in
view.addModel(grep_pdb_rename_chain(f'{structure_output_path}/0.pdb', 'ATOM', 'A'), 'pdb')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/data/lab_website/lab_website_code/pages/1_FlexPose.py", line 132, in grep_pdb_rename_chain
lines = open(pdb_path, 'r').readlines()
^^^^^^^^^^^^^^^^^^^

Hi,

Thanks for your brilliant work and code! I encountered an error when running demo.py:

Could you offer some suggestions?

Bests,

Hi Tiejun,

Thanks for the brilliant work! I took a look at https://github.com/tiejundong/FlexPose/blob/master/FlexPose/utils/prediction.py and find the save_struct function only updates the sidechain angles. Since the alpha carbon coordinates are predicted, have you considered also changing the backbone?

Bests,

Hi, your work is really good and solid! I'm wondering if you can upload your pretraining scripts for protein and ligand encoders? I think your pretraining tasks are quite interesting and informative. Thank you so much!

Great job!
But we would like to use this method locally. Could you provide the trained model? Thank you very much!

Thanks for the work!

I tried to download the weights for local execution, but both at command line
and direct download the speed is slow (< 160 Kb/s) and fails at around 80%.

When trying the web server version, I noticed that aromatics and double-bonded in non-protein compounds tend
often to violate (more or less) co-planar constraints (could be a small setting for minimization in the web server?),
which is not happening in other systems such as DiffDock.
Moreover, big problems occurs when docking symmetric compounds that have multi-aromatics (see the attached example).
Atoms did not stay in place, as it seems that during inference the system cannot distinguish the aromatic ring to which each atom belongs to, and final things result messed up.
Thanks for any support!

Marco

Input_and_server_output.zip

Hi Tiejun,
I noticed that the .chk file seems to be uploaded to your personal website, and the download speed is about 150kb/s. Is there any other faster download method? For example, Google Drive or Baidu Cloud Disk.
Thank you.

I installed Flexpose locally and ran the python demo.py without error, but without any file generation. What could be the reason!