Thank you for such a good work. However, I am feeling confused about those "trained models" in git repo and zenodo.
I used following script and get only NA prediction on zinc_split_1.csv you provided:
nohup python predict.py --folder_path ../inductive_mode/trained_models/bindingdb/inductive/seed12/result/ --test_path ./zinc_data/split_zinc_1.csv --save_dir ../../data/Predictions/ProteomeScreen.csv &
the log says:
99%|█████████▉| 9892/10000 [00:30<00:00, 322.59it/s]
99%|█████████▉| 9925/10000 [00:30<00:00, 322.64it/s]
100%|█████████▉| 9958/10000 [00:30<00:00, 321.67it/s]
100%|█████████▉| 9991/10000 [00:31<00:00, 321.76it/s]
/data/software/miniconda3/envs/graphban/lib/python3.11/site-packages/torch/nn/utils/weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.
warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.")
0
row_average
0 NaN
1 NaN
2 NaN
3 NaN
4 NaN
SMILES ... predicted_value
0 CC(C)(C)c1ccc2occ(CC(=O)Nc3ccccc3F)c2c1\n ... NaN
1 C[C@@h]1CC(Nc2cncc(-c3nncn3C)c2)CC@@HC1\n ... NaN
2 N#Cc1ccc(-c2ccc(OC@@Hc3ccccc3)... ... NaN
3 CCOC(=O)[C@@h]1CCCN(C(=O)c2nc(-c3ccc(C)cc3)n3c... ... NaN
4 N#CC1=C(SCC(=O)Nc2cccc(Cl)c2)N=C([O-])[C@H](C#... ... NaN
I don't understand what's the difference between the model weight you provide in git repo in inductive/transductive folder and those bindingdb_trained_models.zip in zenodo. And I don't know why I get all NA here. Even the model is not that good. I should get some value.
Can you identify the problem? Or can I only use model weights in zenodo when I need to use my own screen data?
Besides, I'm wondering whether your model can be used on reverse DTI situation or other species, like E.coli protein and SMILES? (Just think)
Any help will be very useful, hope this message can reach you asap, thank you very much!!!
Thank you for such a good work. However, I am feeling confused about those "trained models" in git repo and zenodo.
I used following script and get only NA prediction on zinc_split_1.csv you provided:
nohup python predict.py --folder_path ../inductive_mode/trained_models/bindingdb/inductive/seed12/result/ --test_path ./zinc_data/split_zinc_1.csv --save_dir ../../data/Predictions/ProteomeScreen.csv &the log says:
99%|█████████▉| 9892/10000 [00:30<00:00, 322.59it/s]
99%|█████████▉| 9925/10000 [00:30<00:00, 322.64it/s]
100%|█████████▉| 9958/10000 [00:30<00:00, 321.67it/s]
100%|█████████▉| 9991/10000 [00:31<00:00, 321.76it/s]
/data/software/miniconda3/envs/graphban/lib/python3.11/site-packages/torch/nn/utils/weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.
warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.")
0
row_average
0 NaN
1 NaN
2 NaN
3 NaN
4 NaN
SMILES ... predicted_value
0 CC(C)(C)c1ccc2occ(CC(=O)Nc3ccccc3F)c2c1\n ... NaN
1 C[C@@h]1CC(Nc2cncc(-c3nncn3C)c2)CC@@HC1\n ... NaN
2 N#Cc1ccc(-c2ccc(OC@@Hc3ccccc3)... ... NaN
3 CCOC(=O)[C@@h]1CCCN(C(=O)c2nc(-c3ccc(C)cc3)n3c... ... NaN
4 N#CC1=C(SCC(=O)Nc2cccc(Cl)c2)N=C([O-])[C@H](C#... ... NaN
I don't understand what's the difference between the model weight you provide in git repo in inductive/transductive folder and those bindingdb_trained_models.zip in zenodo. And I don't know why I get all NA here. Even the model is not that good. I should get some value.
Can you identify the problem? Or can I only use model weights in zenodo when I need to use my own screen data?
Besides, I'm wondering whether your model can be used on reverse DTI situation or other species, like E.coli protein and SMILES? (Just think)
Any help will be very useful, hope this message can reach you asap, thank you very much!!!