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

1. Using all the descriptors as initially proposed in the Delaney's publication
2. 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

• 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