GFNOrg / gflownet

Generative Flow Networks
MIT License
608 stars 77 forks source link

seh_chembl.csv #17

Open thiviyanT-sony opened 8 months ago

thiviyanT-sony commented 8 months ago

Does the seh_chembl.csv file contain the results from the GFlowNet model output? I am just checking if I am interpreting the CSV correctly.

bengioe commented 8 months ago

Hi @thiviyanT-sony, no, in fact, I'm not entirely sure what this file is doing here. It doesn't seem to be used anywhere in the code. It may be an artifact from the codebase from which this repo was extracted.

thiviyanT-sony commented 8 months ago

Hi @bengioe,

Yes, you are correct. The file is not mentioned in your code.

From the column header, I thought it was part of the results.


[
    'Molecule ChEMBL ID',
    'Molecule Name',
    'Molecule Max Phase',
    'Molecular Weight',
    '#RO5 Violations',
    'AlogP',
    'Compound Key',
    'Smiles',
    'Standard Type',
    'Standard Relation',
    'Standard Value',
    'Standard Units',
    'pChEMBL Value',
    'Data Validity Comment',
    'Comment',
    'Uo Units',
    'Ligand Efficiency BEI',
    'Ligand Efficiency LE',
    'Ligand Efficiency LLE',
    'Ligand Efficiency SEI',
    'Potential Duplicate',
    'Assay ChEMBL ID',
    'Assay Description',
    'Assay Type',
    'BAO Format ID',
    'BAO Label',
    'Assay Organism',
    'Assay Tissue ChEMBL ID',
    'Assay Tissue Name',
    'Assay Cell Type',
    'Assay Subcellular Fraction',
    'Target ChEMBL ID',
    'Target Name',
    'Target Organism',
    'Target Type',
    'Document ChEMBL ID',
    'Source ID',
    'Source Description',
    'Document Journal',
    'Document Year',
    'Cell ChEMBL ID'
]
thiviyanT-sony commented 8 months ago

I am glad I checked this with you. By any chance, do you still have access to the generated molecules from your model? If not, how do I reproduce the GFlowNet results for the molecule generation? Would this be sufficient python mols/gflownet.py?

bengioe commented 8 months ago

I'm afraid not, but yes simply running python gflownet.py should replicate the original model and output the generated molecules. If you're interested in a more robust implementation of GFlowNets on fragments, I'd encourage you to also check out our new repo.