Expanding the chemical space of ILs via conditional VAE

Installation

1. Please follow the installation part here to install moima, which is for VAE generation.
2. Please follow the installation part here to install molops, which is for molecular descriptor calculation.
3. Install tabpfn by pip install tabpfn==2.0.9.
4. Clone this repository:

git clone https://github.com/fate1997/ILGen-ion.git
cd ILGen-ion

Running the workflow

This project including four parts:

1. Train cation and anion scorers and assign scores to PubChem database.

python step1_ion_scorer/ion_scorer.py --ion-type {cation/anion}
python step1_ion_scorer/assign_score.py --ion-type {cation/anion}

2. Train cVAE models to generate cations and anions.

python step2_cVAE/train_cvae.py --ion-type {cation/anion}

3. Train a melting point predicting model.

python step3_melting_point_prediction/train_model.py 

Due to size limit of Github, the model file is not included in this repository. Please download the model file from here and put it in the step3_melting_point_prediction/output directory.

4. Filter ions and generate ILs.

python step4_IL_generation/generate_ils.py

I have also include the pre-trained models under each folder. So you can skip the above process if you just want to sample new ILs

Sampling

You can sample new ILs by running the following command:

python step4_IL_generation/generate_ils.py