This library contains the trained graph neural network model for the prediction of homolytic bond dissociation energies (BDEs) of organic molecules with C, H, N, and O atoms. This package offers a command-line interface to the web-based model predictions at ml.nrel.gov.

The basic interface works as follows, where predict expects a list of SMILES strings of the target molecules

>>> from alfabet import model
>>> model.predict(['CC', 'NCCO'])

  molecule  bond_index bond_type fragment1 fragment2  ...    bde_pred  is_valid
0       CC           0       C-C     [CH3]     [CH3]  ...   90.278282      True
1       CC           1       C-H       [H]    [CH2]C  ...   99.346184      True
2     NCCO           0       C-N   [CH2]CO     [NH2]  ...   89.988495      True
3     NCCO           1       C-C    [CH2]O    [CH2]N  ...   82.122429      True
4     NCCO           2       C-O   [CH2]CN      [OH]  ...   98.250961      True
5     NCCO           3       H-N       [H]   [NH]CCO  ...   99.134750      True
6     NCCO           5       C-H       [H]   N[CH]CO  ...   92.216087      True
7     NCCO           7       C-H       [H]   NC[CH]O  ...   92.562988      True
8     NCCO           9       H-O       [H]    NCC[O]  ...  105.120598      True

The model breaks all single, non-cyclic bonds in the input molecules and calculates their bond dissociation energies. Typical prediction errors are less than 1 kcal/mol. The model is based on Keras and Tensorflow (1.x), and makes heavy use of the neural fingerprint library.

For additional details, see the (upcoming) publication:

• St. John, P.C., Guan, Y., Kim, Y., Kim., S., and Paton, R.S., Prediction of homolytic bond dissociation enthalpies for organic molecules at near chemical accuracy with sub-second computational cost

Installation with conda is recommended, as rdkit can otherwise be difficult to install

$ conda create -n alfabet -c conda-forge python=3.7 rdkit
$ source activate alfabet
$ pip install alfabet