Dataset card: document he2mif/ and he2st/, per-folder licenses

README.md

@@ -1,5 +1,7 @@
 ---
-license: mit
+license: other
+license_name: mixed-per-folder
+license_details: he2mif/ is MIT (MIPHEI-ViT Orion-CRC release); he2st/ is CC BY-NC-SA 4.0 (HEST-1k)
 pretty_name: MuPaD examples
 tags:
   - pathology
@@ -7,6 +9,8 @@ tags:
   - virtual-staining
   - multiplex-immunofluorescence
   - spatial-transcriptomics
+  - xenium
+  - single-cell
 size_categories:
   - n<1K
 ---
@@ -19,8 +23,8 @@ Each folder mirrors the layout of the full dataset, so the released code runs
 on it unchanged: the demos, prediction and scoring, and short training runs.
 
 ```
-he2mif/   H&E -> 16-channel mIF      (5-HE2mIF)   ORION-CRC tiles
-he2st/    H&E -> spatial transcriptomics (6-HE2ST)  HEST-1k Xenium slides
+he2mif/   H&E -> 16-channel mIF               (5-HE2mIF)   ORION-CRC tiles,        393 MB, MIT
+he2st/    H&E -> single-cell transcriptomics  (6-HE2ST)    HEST-1k Xenium slides, 12.4 GB, CC BY-NC-SA 4.0
 ```
 
 Download one folder at a time:
@@ -68,14 +72,59 @@ export ORION_DATA_ROOT=$PWD/examples/he2mif/ORIONCRC_dataset_tile_20x
 ```
 
 Source: the MIPHEI-ViT processed Orion-CRC release, redistributed under its
-MIT license.
+**MIT license**.
 
 > G. Balezo et al. "MIPHEI-ViT: Multiplex Immunofluorescence Prediction from
 > H&E Images using ViT Foundation Models," 2025.
 
 ## he2st/
 
-Two processed HEST-1k Xenium slides for `6-HE2ST`: one training slide and the
-held-out validation slide NCBI920, with their raw HEST files and per-cell MUSK
-features. See `6-HE2ST/README.md` for the environment variables that point the
-code at them.
+Two Xenium breast cancer slides from [HEST-1k](https://github.com/mahmoodlab/HEST),
+processed for H&E → single-cell spatial transcriptomics with MuPaD-512 (`6-HE2ST`).
+
+| slide | split | subject | tiles | cells |
+|---|---|---|---|---|
+| NCBI917 | train | B7 | 945 | 52,688 |
+| NCBI920 | val | B5 | 2,527 | 187,016 |
+
+Both come from *Single-Cell Spatial Transcriptomics of Archival Human Lung and
+Breast Tumor FFPE Samples* (HEST v1.3.1). NCBI920 is a validation slide: the
+model never trains on it, it only selects the checkpoint.
+
+```
+cache/<slide>/              the tile cache (scripts/preprocess_slides.py)
+  he.npy                    (n_tiles, 512, 512, 3) uint8 H&E, 100 µm tiles
+  expr.npy                  (n_cells, 90) float16 log1p expression, tile by tile
+  contours.npy              tile-local cell contours at 512 px
+  contour_offsets.npy       start of each cell's contour
+  cell_seg.npy              each row's index into the xenium_seg polygons
+  tiles.parquet             per tile: origin (full-res px), cell_start, n_cells
+  meta.json                 pixel size, subject, genes, ...
+features/musk/<slide>/
+  emb_cell.npy              (n_cells, 1024) float16 MUSK feature of every cell (he2st.embed_musk)
+HEST1k/                     the raw HEST-1k files these were built from
+  wsis/ st/ xenium_seg/ cells/ metadata/ HEST_v1_3_1.csv (these two rows)
+splits_example.json         train = NCBI917, val = test = NCBI920, for a smoke run
+```
+
+Rows of `expr.npy`, `emb_cell.npy` and every per-cell prediction are the same
+cells in the same order. Tile `t` holds rows `[cell_start, cell_start + n_cells)`.
+
+Point the code at it:
+
+```bash
+export HEST_ROOT=$PWD/examples/he2st/HEST1k HE2ST_CACHE=$PWD/examples/he2st/cache \
+       HE2ST_MUSK_FEATURES=$PWD/examples/he2st/features/musk
+```
+
+The cache can be rebuilt from `HEST1k/` with
+`python scripts/preprocess_slides.py --slides NCBI917,NCBI920` (into an empty
+`HE2ST_CACHE`), which reproduces `cache/` exactly. To score a model on the
+held-out slide, pass `--slides NCBI920` to `scripts/export_mupad_cells.py` and
+`scripts/score.py`. To smoke-test training, set
+`HE2ST_SPLITS=$PWD/examples/he2st/splits_example.json`.
+
+Source: derived from **HEST-1k** (Jaume et al., "HEST-1k: A Dataset for Spatial
+Transcriptomics and Histology Image Analysis", NeurIPS 2024), redistributed
+under its **CC BY-NC-SA 4.0** license: non-commercial use, with attribution,
+and derivatives under the same license.