Dataset card: synthetic, how it was generated, size

README.md

@@ -0,0 +1,82 @@
+---
+license: mit
+pretty_name: spaCR example -- invasion (SYNTHETIC)
+tags:
+- microscopy
+- synthetic
+- spacr
+size_categories:
+- n<1K
+---
+
+# spaCR example data: invasion -- SYNTHETIC
+
+**These images are synthetic. No cell, parasite or antibody was imaged.**
+They exist to show spaCR's **Invasion Assay** module working end to end, and
+say nothing about any real invasion experiment, stain or treatment. No real
+two-colour differential-staining acquisition was available to publish.
+
+Download it from inside spaCR with **Load test data...** in the Invasion
+Assay module, or with `spacr-download invasion`. The settings file ships
+beside the data, so after loading, Run is the next step.
+
+## Size
+
+`spacr-example-invasion.tar`: 106,915,840 bytes (107 MB), an uncompressed tar.
+SHA-256 `881e512447079719d0465d31950bb33ed68ccc099ea7a3024e17f787fbb682d2`.
+
+## How it was generated
+
+`tools/build_invasion_example_dataset.py` in the spaCR repository draws the
+fields with the lattice, object shapes and noise model of `spacr.qt.synthetic`
+(seeded from well and field, so reproducible byte for byte) and writes each
+field as the merged stack spaCR's Mask module produces: four image planes,
+then cell, nucleus and pathogen label planes. **The label planes are the
+objects the generator drew, not a Cellpose segmentation**: Mask with
+Cellpose-SAM did not finish one plate row in 40 minutes on CPU, and the GPU
+was not available. **spaCR's own Measure module** then measured every field,
+so `measurements.db` holds measured intensities, not the generator's
+numbers. `--segment` on the builder runs Mask instead, for a rebuild on a GPU.
+
+The design is the one the Invasion tutorial's synthetic data used: a
+staining-control column and two conditions with a known mixture of invaded
+and attached parasites.
+
+| column | condition | outside stain |
+|---|---|---|
+| c1 | staining control | none on any parasite (the no-primary baseline) |
+| c2 | vehicle | 70% of parasites invaded (no outside stain) |
+| c3 | inhibitor | 35% of parasites invaded |
+
+Rows A-D, 2 fields per well, 512x512 px, 64 host cells
+per field, 75% of them carrying one or two parasites. One attached parasite
+in ten is drawn at a quarter of the outside-stain brightness, so the threshold
+has something to get wrong.
+
+## Channels
+
+| channel | stain | used as |
+|---|---|---|
+| 0 | nucleus | nucleus mask |
+| 1 | cell | cell mask |
+| 2 | total parasite stain (post-permeabilisation) | pathogen mask; `total_channel` |
+| 3 | outside parasite stain (pre-permeabilisation) | `outside_channel` |
+
+## Contents
+
+- `measurements/measurements.db` -- Measure's tables (cell, nucleus,
+  pathogen, cytoplasm).
+- `merged/` -- the 24 merged stacks: four image planes, then the cell,
+  nucleus and pathogen masks (drawn, see above).
+- `ground_truth.csv` -- every parasite the generator drew, with its position,
+  host cell and state. Compare the module's calls against it:
+
+| state | parasites drawn |
+|---|---|
+| attached | 514 |
+| invaded | 617 |
+| no outside stain (staining control) | 582 |
+
+- `settings/invasion_settings.csv` -- the module settings: c1 as the staining
+  baseline, c2 vehicle, c3 inhibitor, channels 3 (outside) and 2 (total).
+  `src` is filled in with the download location when spaCR unpacks it.