This is the code of the bPK score from the paper
Prediction of Small-Molecule Developability Using Large-Scale In Silico ADMET Models
by Beckers et al., Journal of Medicinal Chemistry (2023), https://pubs.acs.org/doi/full/10.1021/acs.jmedchem.3c01083
The bPK score is a deep learning based small molecule scoring approach. The model takes as input the predicted ADMET profile of a compound and is trained on historical pre-clinical development milestones using rank-consistent ordinal regression.
The model is based on PyTorch. Training of the model is described in the Jupyter Notebook train.ipynb. The data of public compounds with annotated milestones is provided in public_milestone_dataset.csv.