boron-nmr-predict

Boron NMR Predict

Use the bundled scripts for deterministic local inference.

Workflow

1. Create and activate a fresh conda environment for this skill.
2. Install Python dependencies from requirements-core.txt and requirements-pyg.txt.
3. Ensure model files exist locally by running scripts/ensure_model.py.
4. Run scripts/predict_boron_nmr.py with a SMILES string and solvent.
5. Return the predicted chemical shift for each boron atom in text form.
6. Generate the labeled PNG image into the user's tmp directory so each ppm value can be matched to B(index) in the structure image.

Input contract

Prefer SMILES input.

Required:

• Molecule SMILES containing at least one boron atom

Optional:

• Solvent name. Supported solvents are:
  • CDCl3
  • C6D6
  • d6-DMSO
  • CD3COCD3
  • CD3CN
  • CD3OD
  • CD2Cl2
  • d8-THF
  • d8-Toluene
  • D2O

Commands

Create a fresh conda environment and install deps:

bash scripts/setup_env.sh

The setup script is portable: it does not assume any machine-specific conda path. It first tries the current shell's conda, then common user-local installs such as ~/miniconda3, ~/anaconda3, and ~/conda.

Manual alternative:

conda create -n boron-nmr-predict python=3.11 -y
conda activate boron-nmr-predict
python -m pip install -r requirements-core.txt
python -m pip install -r requirements-pyg.txt

Download model weights:

python scripts/ensure_model.py

Run prediction:

python scripts/predict_boron_nmr.py \
  --smiles "OB(O)c1ccccc1" \
  --solvent CDCl3 \
  --output-image /tmp/boron_nmr_result.png

Run an example:

bash scripts/run_example.sh

Environment variables

• BORON_NMR_MODEL_REPO: Hugging Face repo id holding the model files
• BORON_NMR_MODEL_DIR: local cache directory for downloaded model files

Defaults:

• model repo: SII-AI4Chem/boron-nmr-predict-model
• model dir: ~/.cache/boron-nmr-predict/models
• image output: user tmp directory such as /tmp/boron_nmr_<id>.png
• device: CPU only