User guide¶
Workflow-oriented walkthroughs. Each page covers one stage of the QSAR workflow and links to the API reference for the details.
Every example in this documentation is executed by the test suite, so the
output shown is what the code actually produces — including the
unflattering numbers. They share one demo dataset, defined in
docs-sphinx/source/demo_data.py and available in every example as
DEMO_SMILES, DEMO_Y and demo_mols.
The demo set is 24 compounds in four substituent series with synthetic pIC50 values, resolving to three Bemis-Murcko scaffolds. It is deliberately small, so every example runs instantly, and deliberately hard: one compound is a planted activity-cliff outlier and the scaffold families are distinct enough that a scaffold split is genuinely difficult. Several pages here show models scoring badly on it. That is the point — an example where everything succeeds teaches nothing about the failure modes these tools exist to detect.
For runnable end-to-end walkthroughs, see the five example notebooks, which together cover every public subpackage.