Spot deconvolution

Shown when: your data has per-cell-type probability columns and spatial coordinates.

The question

Spot-based spatial transcriptomics — Visium most commonly — does not measure single cells: a deconvolution analysis estimates what mixture of cell types each spot contains, giving one row per spot with a probability per cell type:

x	y	Tumor_probability	T_cell_probability	Macrophage_probability
120	340	0.7	0.2	0.1
145	338	0.1	0.1	0.8

BIWT asks whether to expand those spots into individual cells.

Choosing yes

Each spot becomes several cells, allocated in proportion to its probabilities. Your agent-based model gets a plausible cell population rather than a grid of mixture-valued spots.

Choosing yes also means BIWT already knows what the cell types are — they come from the probability column names — so the cluster column screen is skipped.

Deconvolution multiplies your row count — a few thousand spots can become tens of thousands of cells. If that is more than your simulation wants, you can rein it in later at the cell counts screen only if you are not using spatial data; with spatial data the counts follow from the expansion.

Choosing no

BIWT treats the file as ordinary per-row data — one cell per row, at its recorded position. It moves on to the cluster column screen, where you pick a single metadata column holding one label per row; the probability columns are ignored.

This is the right answer when the probability columns are not really a deconvolution — for example, classifier confidence scores where you want the argmax, not a mixture.

Next

Spatial query → if you said yes, or cluster column → if you said no.