fastsolv

This directory contains the fastsolv python package which allows using the trained fastsolv model for solid solubility prediction.

Installation

Run pip install fastsolv to install it from PyPI. Trained model checkpoints will be auto-magically downloaded on your first run of fastsolv.

Requirements

fastsolv is continually tested on all platforms (Windows, MacOS, Linux) with Python versions 3.8 and newer and the latest dependencies. A Graphics Processing Unit (GPU) is optional, but highly recommended for fast predictions. Dependencies are automatically installed when fastsolv is installed with pip - they are fastprop, torch, pandas, and numpy.

Usage

fastsolv is accessible via the command line and as a python module.

• command line: run fastsolv --help for usage instructions or fastsolv demo.csv to run a demo (prints the predicted solubility for aspirin at various temperatures in various solvents, runs in <1 minute + 1st run checkpoint downloading time).
• python module: import the fastsolv predictor with from fastsolv import fastsolv - predictions can then be made by passing a pandas.DataFrame with the columns for solute_smiles, solvent_smiles, and temperature (see demo.ipynb which runs the same predictions as the command line call above).

The CSV files generated as part of the paper directory may also be passed into this predictor.

To manually load fastsolv models and make predictions using torch on your own, adapt the code in fastsolv._module.