PDAC Subtype Agent

1 · Choose your data

Data source
Everything below runs on whichever you pick. A registered dataset is published in biodata-registry with a manifest that already states its quantification, scale and species — nothing to declare. An upload carries no manifest, so you declare and attest to it in the panel that appears.
Dataset
Registered biodata-registry dataset (expression datasets only — DNA/proteomics/single-cell/spatial ids are hidden because this agent cannot analyze them; typing one still gets a worded refusal). Type to search — the list matches on cohort name, author, year and id, so 'moffitt', 'bailey', 'survival' or 'tcga' all narrow it. The Sears/BCC cohort leads, and each cohort's quantification variants are grouped together. Entries marked ⚠ are raw counts: PurIST, GSVA and the charts need per-gene-comparable values, so those refuse with an explanation and the label names the sibling to use instead. Normalized / log-scale inputs (VST, TMM, microarray) run with a caution.

The Compare scores and Heatmap tabs group samples by this cohort's own metadata columns (e.g. tumor_type, a published subtype call). Reading those names means opening the dataset, which on first use downloads it — so it is a button, not something a dropdown does to you silently. You can also just type a column name: every picker accepts free text.

Sample set: all samples (no selection active).

2 · Run an analysis

PurIST basal-like / classical subtyping — on the registered dataset or the upload you chose above.

PurIST is patent-pending, not-for-profit research use only (cite Yeh/Rashid/Moffitt). It needs per-gene-comparable values: a raw-counts source is refused with an explanation (an upload can tick Convert raw counts for this analysis above).