Utilize All Parameters

This is an attempt to create a set of (preferably) small molecules which utilize all parameters in the smirnoff99Frosst force field. Currently, the smirnoff_utilize_all_params.ipynb notebook shows several attempts which utilize a set of molecules from Roche, as well as the first 1,000 and 10,000 molecules (these datasets are all included). However, none of these attempts succeeded in covering the entire set of parameters, each leaving out over 100 parameters.

Hence, we decided to convert the notebook to a script, check_param_coverage.py, which can be run on a computing cluster such as Green Planet to produce parameter ids for a much larger set of molecules. We have done one run so far over a million molecules split across 10 jobs, but 6 of those failed and will be re-run.

Next, we created select_molecules.py to apply a greedy weighted set cover (see here for more details) to find molecules that are small but cover as many parameters as possible.

Finally, we calculated parameters for 1 million molecules in eMolecules and selected a set of molecules from it. We ended up with a set of 71 molecules which cover 299 parameters in smirnoff. The results are stored in the selected folder, which contains the following files:

• chosen.smi - the SMILES strings of the chosen molecules
• param_ids.json - the list of parameter IDs of the chosen molecules
• ignore.json - the list of parameter IDs which were ignored when selecting molecules. These parameters were manually deemed irrelevant.
• remaining.json - the list of parameter IDs which still do not have molecules for them. This list does not include the ones in ignore.json.
• params_by_molecule.json - a mapping from the index of each molecule in chosen.smi to the molecule's SMILES string and the parameter IDs in that molecule. Each ID is annotated with the list of atom indices where the parameter appeared in the molecule (there are multiple lists if the parameter appeared multiple times).

Running

Before running any scripts here, set up the conda environment with

conda env create -f environment.yml

The scripts have the following dependencies:

• openeye-toolkits 2019.4.2
• openforcefield 0.4.0