divelab / airs Goto Github PK

Greetings!

Thank you for releasing the code of the paper 'A Latent Diffusion Model for Protein Structure Generation'. I was wondering if you have planned to release the trained weights and if so, when would you release them?

nothing，i have some mistake。

The MD trajectories almost don't change (see attached video). The QHNet baseline gets 70[μHa] MAE. For the first molecule, the MAE between all 60 Hamiltonians is 45[μHa] (see attached image for all pair differences). This might be caused by the small ~2 attosecond time step. In your reply to reviewer wfG7 you mention a dataset with larger time steps. Any chance you'll publish this within a few days?

Please don't hesitate to let me know if I'm misunderstanding something.

video: https://github.com/divelab/AIRS/assets/8614529/63fd3109-e88f-4b4a-ae4b-eb0a416a87eb

Hello, Thanks for your nice paper! I use main_gen.py to generate ligands for the data in [https://github.com/pengxingang/Pocket2Mol/blob/main/data/test_list.tsv], which is also in the crossdocked2020 dataset, but lots of the ligands I generate are not in the protein pocket (Figure 1 shows the structure of protein pocket with the reference ligand, and Figure 2 shows the structure of protein pocket with the generated ligands) .I would like to ask if these results are correct?

Hi, I meet an error when running conda env create -f environment.yml.

Here is the error message

ERROR: Could not find a version that satisfies the requirement torch-cluster==1.6.0+pt112cu116 (from versions: 0.1.1, 0.2.3, 0.2.4, 1.0.1, 1.0.3, 1.1.1, 1.1.2, 1.1.3, 1.1.4, 1.1.5, 1.2.1, 1.2.2, 1.2.3, 1.2.4, 1.3.0, 1.4.0, 1.4.1, 1.4.2, 1.4.3a1, 1.4.3, 1.4.4, 1.4.5, 1.5.2, 1.5.3, 1.5.4, 1.5.5, 1.5.6, 1.5.7, 1.5.8, 1.5.9, 1.6.0, 1.6.1)

ERROR: No matching distribution found for torch-cluster==1.6.0+pt112cu116

failed

CondaEnvException: Pip failed

Hi, SyMat is really an amazing work. I wonder how to get the score_norm.txt file for a new dataset that I want to train the SyMat on.

Hi. I have a question about irreps in QHNet. In QHNet, it sets self.hidden_irrep = o3.Irreps(f'{self.hs}x0e + {self.hs}x1o + {self.hs}x2e + {self.hs}x3o + {self.hs}x4e') and self.hidden_irrep_base = o3.Irreps(f'{self.hs}x0e + {self.hs}x1e + {self.hs}x2e + {self.hs}x3e + {self.hs}x4e'). These layers are defined as

Hi! I was running the QHNet model using the QH9 dataset these days, but the training loss and validation loss did drop after a few steps. The mae for hamiltonian is 0.05485373 at the begining but it only dropped to 0.01 at the 40,000th step, which is quite weird.

I didn't change any of the code in the original codebase and the dataset was directly downloaded from the Google Drive. I run the main.py directly, and I couldn't find the reason why I can't reproduce the experimental results from the paper.

I wonder if you guys have any comments about that. Any help would be much appreciated!

Should I modify the code using DDP, or is there any provided function for this purpose?

Error happen for the model PotNet

for the version jarvis-tools-2022.9.16, we will meet an error when running
** from jarvis.db.figshare import data
dft_3d = data(dataset='dft_3d')**

Update to the latest version jarvis-tools-2024.4.10 will solve it

Error info :
Obtaining 3D dataset 55k ...
Reference:https://www.nature.com/articles/s41524-020-00440-1
Loading the zipfile...
Traceback (most recent call last):
File "/home/workspace/AIRS/OpenMat/PotNet/main.py", line 23, in
train_prop_model(data, data_root=args.data_root, checkpoint=args.checkpoint, testing=args.testing)
File "/home/workspace/AIRS/OpenMat/PotNet/train_prop.py", line 479, in train_prop_model
result = train_pyg(config, data_root=data_root, file_format=file_format, checkpoint=checkpoint, testing=testing)
File "/home/workspace/AIRS/OpenMat/PotNet/train_prop.py", line 208, in train_pyg
) = get_train_val_loaders(
File "/home/workspace/AIRS/OpenMat/PotNet/data.py", line 507, in get_train_val_loaders
d = jdata(dataset)
File "/home/anaconda3/envs/potnet/lib/python3.9/site-packages/jarvis/db/figshare.py", line 350, in data
dat = get_request_data(js_tag=js_tag, url=url)
File "/home/anaconda3/envs/potnet/lib/python3.9/site-packages/jarvis/db/figshare.py", line 303, in get_request_data
data = json.loads(zipfile.ZipFile(path).read(js_tag))
File "/home/anaconda3/envs/potnet/lib/python3.9/zipfile.py", line 1268, in init
self._RealGetContents()
File "/home/anaconda3/envs/potnet/lib/python3.9/zipfile.py", line 1335, in _RealGetContents
raise BadZipFile("File is not a zip file")
zipfile.BadZipFile: File is not a zip file

I have loaded QH9Stable using this code:
https://github.com/divelab/AIRS/blob/main/OpenDFT/QHBench/QH9/datasets.py

I observe the entries are PyG data objects.

Data(pos=[5, 3], atoms=[5, 1], diagonal_hamiltonian=[5, 14, 14], non_diagonal_hamiltonian=[20, 14, 14], diagonal_hamiltonian_mask=[5, 14, 14], non_diagonal_hamiltonian_mask=[20, 14, 14], edge_index_full=[2, 20])

How do I get the SMILES of a sample?
I want to join (in the database sense) the QH9 records with QM9 records by SMILES.
Thanks!

I tried reproducing the Hamiltonians using PySCF. This gave me a MAE of 2[μHa].

Question 0. Any chance you could release the code to reproduce the dataset? (I likely just made a mistake)

Question 1. Did you see similar errors when creating the dataset?