Installation

Install uv, and from the repo root just type:

uv sync --no-install-package torch-scatter
uv sync --no-build-isolation

Training commands:

Create a .env file in the root of the repo to define the DATA_PATH environment variable. The datasets will be downloaded / processed in this path.

DATA_PATH = your/dataset/path

From the root of the repo run:

Lorenz63

uv run --env-file=.env -- python -m exps.lorenz63.trainer l63 --help

TRP-CAGE (protein-folding)

Before running this experiment, you need to obtain a copy of the data by requesting it at this webpage. Then, follow these steps to preprocess the data:

• Once downloaded, extract the DESRES-Trajectory_2JOF-0-protein.tar.xz file inside the dataset folder as defined in DATA_PATH.
• Copy the topology file exps/trpcage/2JOF-0-protein.pdbat the location DATA_PATH/DESRES-Trajectory_2JOF-0-protein/2JOF-0-protein/2JOF-0-protein.pdb.
• From the root of the repository run the following command:

uv run --env-file=.env -- python -m scripts.to_lmdb --protein-id 2JOF

It is strongly advised to have the DATA_PATH on an SSD. Once the data has been preprocessed, just run

uv run --env-file=.env -- python -m exps.trpcage.trainer trp-cage --help

Calixarene (ligand-binding)

uv run --env-file=.env -- python -m exps.calixarene.trainer G2 --help # G2 ligand
uv run --env-file=.env -- python -m exps.calixarene.trainer G13 --help # G1 + G3 ligands

ENSO

uv run --env-file=.env -- python -m exps.ENSO.trainer ENSO --help