This is an attempt to model the Delany aqueous solubility dataset published in 2004 with contemporary ML approaches using Pycaret package and utilising ensemble stacking of top 3 best performing models.
Two approaches are taken to model the dataset
- Using all the descriptors as initially proposed in the Delaney's publication
- Using VSA and other descriptors as proposed in other publications
git clone https://github.com/aretasg/SolPred.git
cd sol_pred
conda env create -f environment.yml
conda activate solpred| Model | MAE | RMSE | R2 |
|---|---|---|---|
| 1 | 0.4572 | 0.4537 | 0.8886 |
| 2 | 0.4558 | 0.4587 | 0.8874 |
- Both methods performed evidently better then the orignal ESOL and comparably between each other with the former (2) having a slightly better residual plot with less outliers
- For all metrics and other please refer to the end of each jupyter notebook
Both methods are distributed as .pkl files and an example script to run them
- Delaney, 2004
- Avdeef, 2020
MIT license