Python environment for ClassicalGSG

[fwaibl]

Hi,

I would like to use ClassicalGSG to predict logP values of small molecules. However, I am unable to get any non-NaN results. I could imagine that this is because of version mismatches of some Python packages.

I currently have:

• Python 3.8 (with 3.7 I got "ImportError: cannot import name 'cached_property' from 'functools'").
• skorch 0.10.0
• scikit-learn 0.24.2
• torch 1.12.1

When I try to run ClassicalGSG on benzene, I get the following output:

$ python -m LogpPredictor benzene.mol2 benzene.str                                                                                                                                                                                                                                 
/localhome/fwaibl/.conda/envs/classicalgsg/lib/python3.8/site-packages/classicalgsg/molreps_models/gsg.py:64: RuntimeWarning: Precision loss occurred in moment calculation due to catastrophic cancellation. This occurs when the data are nearly identical. Results may be unreliable.                                                                                                                                                                                                                                                                              
  features.append(stats.skew(wavelet_signal, axis=1, bias=False))                                                                                                                                                                                                                  
/localhome/fwaibl/.conda/envs/classicalgsg/lib/python3.8/site-packages/classicalgsg/molreps_models/gsg.py:65: RuntimeWarning: Precision loss occurred in moment calculation due to catastrophic cancellation. This occurs when the data are nearly identical. Results may be unreliable.                                                                                                                                                                                                                                                                              
  features.append(stats.kurtosis(wavelet_signal, axis=1, bias=False))                                                                                                                                                                                                              
/localhome/fwaibl/.conda/envs/classicalgsg/lib/python3.8/site-packages/classicalgsg/molreps_models/gsg.py:84: RuntimeWarning: Precision loss occurred in moment calculation due to catastrophic cancellation. This occurs when the data are nearly identical. Results may be unreliable.                                                                                                                                                                                                                                                                              
  features.append(stats.skew(coefficents, axis=1, bias=False))                                                                                                                                                                                                                     
/localhome/fwaibl/.conda/envs/classicalgsg/lib/python3.8/site-packages/classicalgsg/molreps_models/gsg.py:85: RuntimeWarning: Precision loss occurred in moment calculation due to catastrophic cancellation. This occurs when the data are nearly identical. Results may be unreliable.                                                                                                                                                                                                                                                                              
  features.append(stats.kurtosis(coefficents, axis=1, bias=False))                                                                                                                                                                                                                 
/localhome/fwaibl/.conda/envs/classicalgsg/lib/python3.8/site-packages/sklearn/base.py:310: UserWarning: Trying to unpickle estimator StandardScaler from version 0.23.2 when using version 0.24.2. This might lead to breaking code or invalid results. Use at your own risk.     
  warnings.warn(                                                                                                                                                                                                                                                                   
Predicted logP value is: nan

The same also happens with the MMFF94 version, and also with less symmetric molecules (e.g., pyridine)
Do you know which versions of pytorch etc. I should have installed to use ClassicalGSG?

Best regards,
Franz

[ndonyapour]

Hi @fwaibl,
I'm trying reproduce the error. Would it be possible for you to share the input files (benzene.mol2 and benzene.str)?

Thanks,
Nazanin

[fwaibl]

Hi, I am on holidays this week, but I will send you the files next Monday. For the MMFF94 version, I use the smiles string 'c1ccccc1', with the same error. Best, Franz Am 2. Oktober 2023 00:17:05 GMT+03:00 schrieb Nazanin Donyapour ***@***.***>:

…

Hi @fwaibl, I'm trying reproduce the error. Would it be possible for you to share the input files (benzene.mol2 and benzene.str)? Thanks, Nazanin -- Reply to this email directly or view it on GitHub: #1 (comment) You are receiving this because you were mentioned. Message ID: ***@***.***>

[ndonyapour]

I was able to reproduce the error you are getting. This error is caused by models that were trained with an outdated version of PyTorch. As Python 3.7 is no longer supported, we will need to train our models with higher versions of PyTorch. I am currently working on training new models and will upload them as soon as they are available. Additionally, I will create a tutorial on how to train the ClassicalGSG models.

[alexrd]

Hi @fwaibl , please let us know if this fixes the issue for you!

[fwaibl]

Hi. Thanks for your help. I installed the new version, but now I get the following error message:

FileNotFoundError: [Errno 2] No such file or directory: '/localhome/fwaibl/.conda/envs/classicalgsg/lib/python3.10/site-packages/classicalgsg/pretrained_models/model_4_zfs_MMFF.pkl'

Do I need to re-train the model myself, or should the file be included?

[ndonyapour]

Hi @fwaibl I forgot to push the new trained models. I'll update them.

[fwaibl]

Hi. I just installed the fix_issue_1 branch, and now it seems to work. Thanks for your help!

[alexrd]

Todo: help future-proof this by specifying the versions of different packages in setup.py