deepchem

1# DeepChem
2 
3## When to use
4 
5Use for molecular property prediction, SMILES/graph featurization, MoleculeNet
6benchmarks and explicitly configured encoder transfer. The workflow also covers
7DeepChem materials and sequence adapters when their data/model contracts are met.
8 
9Targets **DeepChem 2.8.0**, still the stable PyPI release at review. Nightly 2.8.1
10builds have additional APIs and different constraints; do not mix `latest` docs
11with a stable installation. The tested CPU stack and optional-backend limitations
12are in [references/review.md](references/review.md).
13 
14## Workflow
15 
161. Define compound identity, assay conditions, target units and missing labels.
17   Parse SMILES and retain a record of rejected rows. Audit duplicate structures
18   and repeated measurements before splitting.
192. Match the representation to the model using the table below. Fit a numeric
20   baseline first; dataset size alone does not select the best architecture.
213. Choose a scaffold, temporal or grouped holdout for the deployment question.
22   Preserve exact split IDs, inspect scaffold overlap and per-task class support,
23   and reject empty splits. Scaffold splitting does not eliminate all leakage.
244. Fit preprocessing on training data only. Preserve `dataset.w` masks. Normalize
25   continuous targets only, and invert those transforms for metrics/predictions.
265. Select model settings/epochs on validation data and evaluate the final holdout
27…