Requirements

The code is built upon chemprop-v1 and descriptastorus. The required versions are provided in this repository. All credit for the chemprop and descriptastorus packages goes to their respective authors. Our models needs the following requirements:

• python 3.9
• pytorch 1.10.1
• rdkit 2023.03.3
• numpy 1.23.1
• pandas 1.4.3 The full list of requirements can be found in the requirements.txt file. The code was tested on Ubuntu 20.04.3 LTS.

Installation

1. install conda/miniconda (https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html)
2. create a new conda environment: conda create -n nf10k python=3.9
3. activate the environment: conda activate nf10k
4. install the required packages: pip install -r requirements.txt The installation process (without conda/miniconda) will take around ~5 minutes.

Running predictions

Solute rejection prediction can be made by running the prediction.py script. The script requires the following arguments:

• --smiles <"SMILES string of the molecule">
• --membrane <membrane> the script only accepts: DM300, GMT-oNF-2, PBI, NF90, PMS600, SM122, NF270
• --solvent <solvent_name> the script only accepts: Water, Toluene, "Methyl tetrahydrofuran", Methanol, Ethanol, "Dimethyl formamide", Acetonitrile, Acetone, "Ethyl acetate"
• --configuration <configuration> the script only accepts: OSN, "Loose NF", NF For example, to predict the rejection of paracematol in methanol in Duramem 300 membrane, run this command:
• python prediction.py --smiles "CC(=O)Oc1ccccc1C(=O)O" --membrane DM300 --solvent Methanol --configuration OSN Expected output: {'smiles': 'CC(=O)Oc1ccccc1C(=O)O', 'membrane': 'DM300', 'solvent': 'Methanol', 'configuration': 'OSN', 'rejection': 0.841'} where the rejection is the predicted rejection of the molecule in the given conditions. There is also user_predictions.csv file that contains the predictions of the last run. Re-running the script will overwrite the previous results. A single-solute run takes around ~10 seconds to finish.

Recreating the results of the paper

The results of the paper can be recreated by running the run_paper_predictions.py script. The script will run the predictions for all the molecules. The results will be saved in the results/paper_predictions.csv. The script takes between 2-5 minutes to finish.

License

This code is released under the MIT License.