Generate OptimizedConfiguration for GAMESS using ANI

Optimizes the geometry of a molecule using the ANI neural potential (ani2x, trained upon wB97x/6-31G* for C, H, N, O, S, F, and Cl)) and adds the optimized geometry to an input file for GAMESS (e.g. nvcr.io/hpc/gamess:17.09-r2-libcchem).

Quick links

• Getting started
• Development
• Help and Support
• Contributing
• License

Getting started

1. Get access to an environment hosting a deployment of the Simulation Toolkit for Scientific Discovery (ST4SD).
2. Push the experiment to your ST4SD deployment.
   1. You can find the JSON representation of the parameterised package in 'parameterised-packages/configuration-generator-ani.json'.
3. Start your experiment
4. Download the outputs of your experiment run after your run terminates

Adding the experiment to ST4SD

This will add a basic parameterised package for this experiment called configuration-generator-ani

api.api_experiment_push({"base": {"packages": [{"config": {"manifestPath": "manifest.yaml", "path": "ani-surrogate.yaml"}, "dependencies": {"imageRegistries": []}, "name": "main", "source": {"git": {"location": {"tag": "1.0.0", "url": "https://github.com/st4sd/configuration-generator-ani.git"}}}}]}, "metadata": {"package": {"description": "Surrogate that optimizes the geometry of a molecule using the ANI neural potential (ani2x, functional: vWB97x) and adds it to a GAMESS molecule.inp file", "keywords": ["smiles", "computational chemistry", "geometry-optimization", "gamess", "surrogate"], "maintainer": "https://github.com/Jammyzx1", "name": "configuration-generator-ani", "tags": ["1.0.0"]}}, "parameterisation": {"executionOptions": {"data": [], "platform": [], "runtime": {"args": [], "resources": {}}, "variables": [{"name": "molecule_index"}, {"name": "n_conformers"}, {"name": "max_iterations"}]}, "presets": {"data": [], "environmentVariables": [], "platform": "openshift", "runtime": {"args": ["--registerWorkflow=yes"], "resources": {}}, "variables": [{"name": "ani_model", "value": "ani2x"}, {"name": "optimizer", "value": "bfgs"}, {"name": "functional", "value": "wB97X"}]}}})

You can create your own parameterised packages for these experiments. See here for more information about how to do this. This repo contains files with the above json in easy to edit form you can use as a basis.

Experiment inputs

The experiment requires an input CSV file, called input_smiles.csv, with columns label and smiles. The label column should contain unique integers (e.g. the row number). The smiles column should contain the SMILE representation of the input molecules. The file can contain other columns - these will be ignored.

Example:

label,smiles
0,O=S(=O)([O-])c1c(C(F)(F)F)cc(C(F)(F)F)cc1C(F)(F)F.Cc1cc(OC(C)(C)C)cc(C)c1[S+](c1ccccc1)c1ccccc1
1,O=S(=O)([O-])c1ccc(C(F)(F)F)cc1C(F)(F)F.Cc1ccc([S+](c2ccccc2)c2ccccc2)c(C)c1

An example payload for the experiment can be created as follows:

smiles_data = pd.read_csv('smiles_data.csv')[['label', 'SMILES']]
payload = {
    "inputs": [{
       "content": smiles_data.to_csv(index=False),
       "filename": "input_smiles.csv"
   }]
}

Development

1. Fork this repository. You will find a virtual experiments in this repository.
2. Modify the code
3. Push your code to your forked github repository. Then follow the getting started instructions above.

Note: Remember to update your parameterised package payload so that it points to your forked GitHub repository.

Help and Support

Please feel free to reach out to one of the maintainers listed in the MAINTAINERS.md page.

Contributing

We always welcome external contributions. Please see our guidance for details on how to do so.

License

This project is licensed under the Apache 2.0 license. Please see details here.