Deepchem molgraph featurization

Purpose

Create deterministic DeepChem molecular graph features from a small SMILES table using the repo-managed chemtools prefix.

When to use

• You need a lightweight DeepChem starter without training a model.
• You want to inspect graph sizes and node-feature dimensions before modeling.

When not to use

• You need trained DeepChem models requiring TensorFlow or PyTorch.
• You need dataset download or benchmark automation.

Inputs

• A TSV file with molecule_id and smiles columns.

Outputs

• A JSON summary with graph dimensions for each molecule.

Requirements

• slurm/envs/chemtools
• DeepChem and RDKit in that prefix

Procedure

1. Run slurm/envs/chemtools/bin/python skills/computational-chemistry-and-molecular-simulation/deepchem-molgraph-featurization/scripts/featurize_molecules.py --input skills/computational-chemistry-and-molecular-simulation/deepchem-molgraph-featurization/examples/molecules.tsv --out scratch/deepchem/featurization.json.
2. Inspect node counts, edge counts, and feature dimensions.
3. Use the feature summary as a preflight step before larger molecular ML runs.

Validation

• The script exits successfully.
• Each molecule yields graph metadata.
• Node and edge counts are positive for valid molecules.

Failure modes and fixes

• Missing optional ML backends: this skill only requires the featurizer path, not torch or tensorflow.
• Invalid SMILES: correct the input row before featurization.

Provenance

• DeepChem documentation: https://deepchem.readthedocs.io/en/latest/

Related skills

• rdkit-molecular-descriptors