Save xenium version in raw-ist

scripts/create_resources/combine/process_datasets_xenium_nebius.sh

@@ -129,6 +129,17 @@ param_list:
     dataset_description: "Xenium V1 FFPE Human Breast IDC + 2021 Wu scRNAseq"
     dataset_organism: "homo_sapiens"
 
+  - id: "2026_10x_human_breast_cancer_atera_combined"
+    input_sp: "$input_dir/10x_atera/2026_10x_human_breast_cancer_atera/dataset.zarr"
+    input_sc: "$input_dir/wu_human_breast_cancer_sc/2021Wu_human_breast_cancer_sc/dataset.h5ad"
+    dataset_id: "2026_10x_human_breast_cancer_atera_combined"
+    dataset_name: "Human breast cancer combined 2026 10x Atera WTA 2021 Wu scRNAseq"
+    dataset_url: "https://www.10xgenomics.com/datasets/atera-wta-ffpe-human-breast-cancer"
+    dataset_reference: "https://doi.org/10.1038/s41588-021-00911-1"
+    dataset_summary: "Atera WTA FFPE Human Breast Cancer (DCIS Grade 3) + 2021 Wu scRNAseq"
+    dataset_description: "Atera WTA FFPE Human Breast Cancer (DCIS Grade 3) + 2021 Wu scRNAseq"
+    dataset_organism: "homo_sapiens"
+
 output_sc: "\$id/output_sc.h5ad"
 output_sp: "\$id/output_sp.zarr"
 output_state: "\$id/state.yaml"

src/datasets/loaders/tenx_atera/script.py

@@ -1,6 +1,7 @@
 ## code author: Florian Heyl
 import shutil
 import os
+import json
 import zipfile
 import tempfile
 from pathlib import Path
@@ -140,6 +141,18 @@ def extract_zip(input_zip: Path, output_dir: Path, strip_root: bool = False):
 for key, value in new_uns.items():
     sdata.tables["table"].uns[key] = value
 
+# Preserve the Xenium analysis software version (from the raw experiment.xenium) so
+# downstream components can recover it instead of hardcoding — e.g. segger reads it to
+# select its v1 vs v2+ Xenium loader. Best-effort: skip if absent/unreadable.
+try:
+    with open(input_extracted / "experiment.xenium") as f:
+        xenium_sw_version = json.load(f).get("analysis_sw_version")
+    if xenium_sw_version:
+        sdata.tables["table"].uns["xenium_analysis_sw_version"] = xenium_sw_version
+        print(f"Xenium analysis_sw_version: {xenium_sw_version}", flush=True)
+except (OSError, ValueError) as e:
+    print(f"(no experiment.xenium analysis_sw_version: {e})", flush=True)
+
 # Force a regular chunk grid so the written store is readable by spatialdata's
 # ome-zarr reader (see rechunk_uniform). Do NOT enable array.rectilinear_chunks:
 # that only permits writing the unreadable rectilinear grids we are avoiding.

src/datasets/loaders/tenx_xenium/script.py

@@ -3,6 +3,7 @@
 from spatialdata_io import xenium
 import shutil
 import os
+import json
 import zipfile
 import tempfile
 
@@ -97,6 +98,18 @@
     for key, value in new_uns.items():
         sdata.tables["table"].uns[key] = value
 
+    # Preserve the Xenium analysis software version (from the raw experiment.xenium)
+    # so downstream components can recover it instead of hardcoding — e.g. segger reads
+    # it to select its v1 vs v2+ Xenium loader. Best-effort: skip if absent/unreadable.
+    try:
+        with open(os.path.join(par_input, "experiment.xenium")) as f:
+            xenium_sw_version = json.load(f).get("analysis_sw_version")
+        if xenium_sw_version:
+            sdata.tables["table"].uns["xenium_analysis_sw_version"] = xenium_sw_version
+            print(f"Xenium analysis_sw_version: {xenium_sw_version}", flush=True)
+    except (OSError, ValueError) as e:
+        print(f"(no experiment.xenium analysis_sw_version: {e})", flush=True)
+
     print(f"Output: {sdata}", flush=True)
 
     print(f"Writing to '{par['output']}'", flush=True)

src/methods_transcript_assignment/segger/config.vsh.yaml

@@ -65,13 +65,16 @@ engines:
       - type: python
         github:
           - openproblems-bio/core#subdirectory=packages/python/openproblems
-      # torch for the GNN. Standard PyPI cu124 wheel (numpy-2 compatible) coexists
-      # with the RAPIDS cuda-12 libs. Unlike the NGC custom torch, torch_scatter has
-      # a PREBUILT wheel for this torch (the pyg index) — no CUDA compile needed.
+      # torch (+ torchvision, which segger's data_module imports) for the GNN.
+      # Standard PyPI cu124 wheels (numpy-2 compatible) coexist with the RAPIDS
+      # cuda-12 libs. torchvision must match torch (2.6.0 <-> 0.21.0). Unlike the NGC
+      # custom torch, torch_scatter has a PREBUILT wheel for this torch (pyg index) —
+      # no CUDA compile needed. typer + torch_geometric + lightning are pure Python
+      # (segger imports all three but declares none of them).
       - type: docker
         run:
-          - pip install --no-cache-dir torch==2.6.0 --index-url https://download.pytorch.org/whl/cu124
-          - pip install --no-cache-dir torch_geometric lightning
+          - pip install --no-cache-dir torch==2.6.0 torchvision==0.21.0 --index-url https://download.pytorch.org/whl/cu124
+          - pip install --no-cache-dir torch_geometric lightning typer
           - pip install --no-cache-dir torch_scatter -f https://data.pyg.org/whl/torch-2.6.0+cu124.html
       # I/O stack. spatialdata pinned <0.8 (resolves to 0.7.1): 0.7.2+ require
       # dask>=2026.x, but RAPIDS 25.04 hard-pins dask==2025.2.0. 0.7.1 accepts dask

src/methods_transcript_assignment/segger/script.py

@@ -1,4 +1,5 @@
 import os
+import json
 import shutil
 import subprocess
 from pathlib import Path
@@ -154,6 +155,26 @@
 tx_out.to_parquet(XENIUM_DIR / "transcripts.parquet", index=False)
 del transcripts, y_coords, x_coords, label_image
 
+# segger 0.1.0 auto-detects the platform from marker files: its Xenium loader needs an
+# `experiment.xenium` JSON and reads only `analysis_sw_version` from it, which selects its
+# v1 ("-1") vs v2+ ("UNASSIGNED") loader. We always write the "UNASSIGNED" sentinel above,
+# so we need the v2+ loader (version >= 2). Reuse the real version the Xenium dataset
+# loader stashed in the metadata table's uns when it is v2+; otherwise (v1 source, missing,
+# or a non-Xenium dataset wrapped here as Xenium-layout) fall back to a v2+ default.
+DEFAULT_SW_VERSION = "xenium-3.0.0"
+sw_version = DEFAULT_SW_VERSION
+try:
+    src_version = sdata.tables["metadata"].uns.get("xenium_analysis_sw_version")
+    if src_version and int(str(src_version).split("-")[-1].split(".")[0]) >= 2:
+        sw_version = str(src_version)
+    elif src_version:
+        print(f"Source Xenium version '{src_version}' is <2, but we write the "
+              f"UNASSIGNED sentinel; using {sw_version} instead.", flush=True)
+except (KeyError, AttributeError, ValueError, IndexError):
+    pass
+print(f"Writing experiment.xenium with analysis_sw_version={sw_version}", flush=True)
+(XENIUM_DIR / "experiment.xenium").write_text(json.dumps({"analysis_sw_version": sw_version}))
+
 ################
 # Run segger   #
 ################