Designing new molecules as fast as possible with AI and simulation.

• Documentation: exalearn.github.io/ExaMol/
• Source Code: github.com/exalearn/ExaMol

First clone this repository to the computer you'll use to design molecules:

git clone https://github.com/exalearn/ExaMol.git

You can then build the entire environment with Anaconda:

cd ExaMol
conda env create --file envs/environment-cpu.yml --force

The above command builds the environment for a commodity CPU. We will provide environments compatible with supercomputers over time.

ExaMol deploys a computational workflow following a specification which that contains information including:

• what is molecular properties are already known,
• what are the molecules we could search,
• the types of computations being performed,
• how to search over them (e.g., how to schedule tasks, which active learning strategy, informed by which ML model),
• and the resources over which computations are deployed.

An example which performs a greedy search using xTB would look something like

recipe = RedoxEnergy(charge=1, compute_config='xtb')  # What we're trying to optimize
spec = ExaMolSpecification(
    database='training-data.json',
    recipe=[recipe],
    search_space='search_space.smi',
    selector=GreedySelector(n_to_select=8, maximize=True),
    simulator=ASESimulator(scratch_dir='/tmp'),
    scorer=RDKitScorer(recipe),
    models=[[KNeighborsRegressor()]],
    num_to_run=8,
    thinker=SingleStepThinker,
    compute_config=config,
    run_dir='run'
)

The full example is shown in our examples directory

The redoxmer example we allude to above can be run from the command line and requires only a few minutes to complete:

examol run examples/redoxmers/spec.py:spec

The above command simply tells ExaMol the name of the file containing the run spec, and the name of the spec within that file. ExaMol will then run it for you.

Initial development of ExaMol was funded jointly by the ExaLearn Co-design Center of the Department of Energy Exascale Computing Project and the Joint Center for Energy Storage Research (JCESR), an Energy Innovation Hub funded by the Department of Energy.