- Convolutional Networks on Graphs for Learning Molecular Fingerprints by David Duvenaud et al.
This software is implemented by DGL + PyTorch. Note that this version cannot use mini-batch nor GPU training as it is.
Prerequisites include below:
After installing them, do commands below:
git clone https://github.com/Masatsugar/neural_fingerprint.git
python setup.py install- minimum usage
from neural_fingerprint import NFPRegressor
import rdkit.Chem as Chem
from rdkit.Chem import Descriptors
import numpy as np
mols = [Chem.MolFromSmiles(smi) for smi in ['CC', 'CCC', 'COC', 'C(=O)CC', 'CNC']]
logP = np.array([Descriptors.MolLogP(mol) for mol in mols])
nfp = NFPRegressor()
nfp.fit(mols, logP, epochs=10)
preds, fps = nfp.predict(mols, return_fps=True)