Errors when run pytest tests #56
Comments
|
Hi, @mattragoza I run the code: "python3 generate.py config/generate.config". There is also an error: This seems to be related to mismatch of nn.linear. But I don't know how to solve it. Thanks! |
|
Hm, I have not seen this error. This post on the pytorch forum suggests that it might be resolved by reducing the batch size. I don't think it's a dimension mismatch as these tests should all be passing if you are up to date with the master branch. |
|
How much GPU memory does your card have? |
I tried to reduce the batch size to 1, but the error is still there. My GPU memory is 32G. |
|
If you run the code on the CPU, it may give a more informative error message. I just pushed a commit that lets you control this by setting the config option Can you try that and post the error message? |
Did you mean add It still reported the same error. |
Yes.
Did you |
When I git pull via 'git pull --rebase', it works. But there is still some warnings. Part of the output is below: I use a protein with no ligand. As said in the #44 , you said I can use any ligand (use tests/input/O_0_0_0.sdf) in prior sampling. And I also don't know where the binding pocket is. I just only use the protein and O_0_0_0.sdf in the file, like this: The output is all H2O: I am a beginner, so there are some issues that need to be solved: (4) If (3) yes, why all my results are the same: H2O? |
You can ignore OpenBabel warnings.
If the code works when you use
The configuration setting
Yes, you should be able to do prior sampling with an empty/dummy input ligand. However, note that the receptor and ligand grids are centered on the input ligand file, and generated ligands will be centered there accordingly. So in this case, the grid will be centered at the origin, which might result in the receptor grid containing an arbitrary location on the receptor structure. It might not contain any receptor structure at all, depending on the receptor structure coordinates.
This is probably because the grid is not centered correctly and contains very little of the receptor structure, so the model is generated a ligand based on mostly empty space. It's ok if you do not have an input ligand, but you at least need to decide where the grid will be centered, by changing the coordinates on the dummy atom in the file We have not tested this generative model on receptors with unknown binding pockets, so I can't say what the result will be or what the best approach is for discovering potential binding pockets. But at the very least you need to consider the fact that the entire receptor structure is unlikely to fit into the grid, so you need to decide where to center the grid. You could possible try scanning the entire receptor structure by generating many dummy ligand files centered on random locations or a uniform grid spanning the coordinate domain of the receptor structure, then generating ligands at each location and assessing their gnina score. I don't know how effective this would be but it's one possible approach!
|
|
Hi @mattragoza I guess I can use dock model, such as autodock, to determine ligand coordinates and the grids. right? |
|
Yes, that would work too. |
|
Hi @mattragoza I still have some issues to solve it. （3）Before run the LiGAN, did I need to do something else with the protein or the ligand, such as adding hydrogens and charge? |
Hi, @mattragoza I installed LiGAN successfully. But when I run pytest tests, there is a lot of errors occurred:
I don't know if it's a cuda issue. Thanks!
The text was updated successfully, but these errors were encountered: