A Hybrid Framework For Uncertainty Quantification in Deep Learning-Based QSAR Regression Modeling

---

Requirements

• python == 3.7.0
• numpy
• pandas
• scipy
• scikit-learn
• pytorch
• tqdm
• tap (pip install typed-argument-parser)
• rdkit

Building of D-MPNN models are based on chemprop package (https://github.com/chemprop/chemprop).

Datasets

All datasets are stored in ~/data/dataset
Splitting indexes for three different splitting strategies are stored in ~/data/ivit, ~/data/ivot and ~/data/ovot

Usage

To produce 5-fold cross-validation results on 24 datasets using the D-MPNN model, please run:

$ python script/produce_results.py SPLITTING_TYPE

For example:

$ python script/produce_results.py ivit

Here SPLITTING_TYPE should be one of ivit,ivot and ovot

All results will be stored in ~/results