Closed odek53r closed 6 years ago
Hi,
In order to train the model with synthetic graphs, you can use https://github.com/Networks-Learning/nevae/blob/master/generate_erdos_renyi.py code and get a generated set of graphs.
You can use
python generate_erdos_renyi.py --help
to know the parameters.
Also to generate molecular graphs for ZINC data set you can download the clean druglike molecules from
http://zinc.docking.org/subsets/clean-drug-like
Then you can use the .mol2 files and .sml files and checkout node_label
branch.
Then use the following command
python molecular_graph_conversion.py .sml .mol2
Similarly, download the QM9 data from
https://figshare.com/collections/Quantum_chemistry_structures_and_properties_of_134_kilo_molecules/978904
and get the smiles, convert them to mol2 file using rdkit and use the code to convert it to a molecular graph to be used by networkx module.
For example:
The molecular graph corresponding to the following smile string:
CC(C)(O)CC1CCNCC1
is
0 1 {"weight":1}
0 11 {"weight":1}
0 12 {"weight":1}
0 13 {"weight":1}
1 2 {"weight":1}
1 3 {"weight":1}
1 10 {"weight":1}
2 14 {"weight":1}
2 15 {"weight":1}
2 16 {"weight":1}
3 4 {"weight":1}
3 17 {"weight":1}
3 18 {"weight":1}
4 9 {"weight":1}
4 5 {"weight":1}
4 29 {"weight":1}
5 6 {"weight":1}
5 19 {"weight":1}
5 20 {"weight":1}
6 7 {"weight":1}
6 21 {"weight":1}
6 22 {"weight":1}
7 8 {"weight":1}
7 28 {"weight":1}
8 9 {"weight":1}
8 23 {"weight":1}
8 24 {"weight":1}
9 25 {"weight":1}
9 26 {"weight":1}
10 27 {"weight":1}
Hope this helps.
Thank you for the prompt reply. I convert mol2 to graph successfully.
How can I get the training data? It seems there doesn't have introduction about data format or data examples for running the program. I'll appreciate for your help.