Skip to content

Latest commit

 

History

History
58 lines (48 loc) · 2.69 KB

pubchem.md

File metadata and controls

58 lines (48 loc) · 2.69 KB
Raw

PubChem

PubChem18 (used in the paper)

For the PubChem dataset used in the paper run the following commands, which downloads, unzips and deletes the zip-file:

wget -N -r https://cloud.ml.jku.at/s/2ybfLRXWSYb4DZN/download -O pubchem18.zip
unzip pubchem18.zip; rm pubchem18.zip

PubChem23 / up-to-date PubChem

To setup the PubChem pretraining dataset with the latest version call:

python clamp/dataset/prep_pubchem.py --data_dir=./data/pubchem23/

This downloads the PubChem-database and preprocesses it. You may manually delete the raw PubChem Folder (data_dir/ftp.ncbi.nlm.nih.gov) after running this command. Get some coffee, this will take quite some time ;)

For reproducibility, data is also available through

wget -N -r https://cloud.ml.jku.at/s/fi83oGMN2KTbsNQ/download -O pubchem23.zip
unzip pubchem23.zip
rm pubchem23.zip

Satistics

PubChem23 (2023-02) PubChem18 PubChem HTS FSMOL (v1)
# measurements 245,234,081 223,219,241 143,886,653 501,366
# compounds 3,475,417 2,120,811 715,231 240,465
# assays 521,603 21,002 582 5,135
Mean # compounds / assay 462.69 10,628.48 247,227.93 96.32
Median # compounds / assay 2.00 35.00 304,804.00 46.00
% active 2.58 1.51 0.70 46.48
Mean % active per assay 80.57 79.46 1.04 47.17
Median % active per assay 100.00 100.00 0.42 48.84
Source PubChem 2023 PubChem 2018 PubChem 2018 ChEMBL27
% of assays with only one class 95.54 74.54 1.37 0.00
% density 0.01 0.50 34.57 0.04

Compound and Assay Encodings

To compute the compound encodings as input for your model run

python clamp/dataset/encode_compound.py \
--compounds=./data/pubchem23/compound_names.parquet \
--compound2smiles=./data/pubchem23/compound_smiles.parquet \
--fp_type=morganc+rdkc --fp_size=8096

To compute the assay encodings as input for your model run

python clamp/dataset/encode_assay.py --encoding=clip

or use --encoding=lsa.