deepchem

deepchem is an agent skill for building molecular property prediction and MoleculeNet workflows with DeepChem. It covers SMILES and graph featurisation, scaffold or grouped holdouts that keep related molecules together, masked labels, MoleculeNet benchmarks, and explicitly configured encoder transfer. It also covers DeepChem's materials and sequence adapters, as long as their data and model contracts are met. The skill targets DeepChem 2.8.0, which is still the stable PyPI release at the time of its review. Nightly 2.8.1 builds have additional APIs and different constraints, so the skill advises not to mix the latest documentation with a stable installation. The tested CPU stack and the limitations of optional backends are recorded in a review file. Use it when you are predicting molecular properties from structures, running standard benchmarks, or setting up a model that must be evaluated on held-out chemical groups rather than random splits. It is part of the k-dense-ai scientific agent skills repository on GitHub, which has about 49,600 stars. The listing shows it as free, and it has about 1,800 installs.…