Fixes

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"

scripts/run_benchmark/run_full_local.sh

@@ -64,9 +64,11 @@ celltype_annotation_methods:
   # - rctd
 expression_correction_methods:
   - no_correction
-  # - gene_efficiency_correction
   # - resolvi_correction
   # - split
+gene_efficiency_correction_methods:
+  - no_correction
+  # - gene_efficiency_correction
 method_parameters_yaml: /tmp/method_params.yaml
 HERE
 

scripts/run_benchmark/run_full_seqeracloud.sh

@@ -56,9 +56,11 @@ celltype_annotation_methods:
   - rctd
 expression_correction_methods:
   - no_correction
-  - gene_efficiency_correction
   - resolvi_correction
   - split
+gene_efficiency_correction_methods:
+  - no_correction
+  - gene_efficiency_correction
 method_parameters_yaml: /tmp/method_params.yaml
 HERE
 

scripts/run_benchmark/run_test_local.sh

@@ -59,9 +59,11 @@ celltype_annotation_methods:
   # - rctd
 expression_correction_methods:
   - no_correction
-  # - gene_efficiency_correction
   # - resolvi_correction
   # - split
+gene_efficiency_correction_methods:
+  - no_correction
+  # - gene_efficiency_correction
 method_parameters_yaml: /tmp/method_params.yaml
 HERE
 

scripts/run_benchmark/run_test_segger_nebius.sh

@@ -44,6 +44,9 @@ celltype_annotation_methods:
   - tacco
 expression_correction_methods:
   - no_correction
+gene_efficiency_correction_methods:
+  - no_correction
+  # - gene_efficiency_correction
 HERE
 
 # Write the parameters to file (input_states version, NOTE: enable `-entry_name auto` for this)

scripts/run_benchmark/run_test_seqeracloud.sh

@@ -55,9 +55,11 @@ celltype_annotation_methods:
   - rctd
 expression_correction_methods:
   - no_correction
-  - gene_efficiency_correction
   - resolvi_correction
   - split
+gene_efficiency_correction_methods:
+  - no_correction
+  - gene_efficiency_correction
 #method_parameters_yaml: /tmp/method_params.yaml
 HERE
 

src/api/comp_method_gene_efficiency_correction.yaml

@@ -0,0 +1,31 @@
+namespace: methods_gene_efficiency_correction
+info:
+  type: method
+  subtype: method_gene_efficiency_correction
+  type_info:
+    label: Gene efficiency correction
+    summary: Correcting per-gene detection efficiency in spatial data
+    description: >-
+      A gene efficiency correction method corrects for per-gene detection efficiency in the
+      (expression-)corrected spatial counts. It runs as a separate stage after expression
+      correction, consuming and producing the corrected-counts format.
+arguments:
+  - name: --input
+    required: true
+    direction: input
+    __merge__: /src/api/file_spatial_corrected_counts.yaml
+  - name: --input_scrnaseq_reference
+    required: false
+    direction: input
+    __merge__: /src/api/file_scrnaseq_reference.yaml
+  - name: --output
+    required: true
+    direction: output
+    __merge__: /src/api/file_spatial_corrected_counts.yaml
+test_resources:
+  - path: /resources_test/task_ist_preprocessing/mouse_brain_combined
+    dest: resources_test/task_ist_preprocessing/mouse_brain_combined
+  - type: python_script
+    path: /common/component_tests/run_and_check_output.py
+  - type: python_script
+    path: /common/component_tests/check_config.py

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_cell_type_annotation/moscot/script.py

@@ -75,11 +75,13 @@
 adata_sp.X = adata_sp.layers["normalized"]
 adata_sp.obsm["spatial"] = adata_sp.obs[["centroid_x", "centroid_y"]].to_numpy()
 
+sc.pp.pca(adata_sc, n_comps=50)   # X is the normalized layer set above
+
 # Define mapping problem
 mp = MappingProblem(adata_sc=adata_sc, adata_sp=adata_sp)
 
 mp = mp.prepare(
-    sc_attr={"attr": "layers", "key": "normalized"},
+    sc_attr={"attr": "obsm", "key": "X_pca"},   # <-- 50-dim, not raw genes
     xy_callback="local-pca",
 )
 

src/methods_expression_correction/gene_efficiency_correction/config.vsh.yaml → src/methods_gene_efficiency_correction/gene_efficiency_correction/config.vsh.yaml

@@ -1,4 +1,4 @@
-__merge__: /src/api/comp_method_expression_correction.yaml
+__merge__: /src/api/comp_method_gene_efficiency_correction.yaml
 
 name: gene_efficiency_correction
 label: "Gene Efficiency Correction"
@@ -27,7 +27,7 @@ resources:
 engines:
   - type: docker
     image: openproblems/base_python:1
-    __merge__: 
+    __merge__:
       - /src/base/setup_txsim_partial.yaml
     setup:
         - type: python
@@ -38,4 +38,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [ midtime, lowcpu, midmem ]
+      label: [ midtime, lowcpu, midmem ]

src/methods_expression_correction/gene_efficiency_correction/script.py → src/methods_gene_efficiency_correction/gene_efficiency_correction/script.py

@@ -5,29 +5,33 @@
 # Note: this section is auto-generated by viash at runtime. To edit it, make changes
 # in config.vsh.yaml and then run `viash config inject config.vsh.yaml`.
 par = {
-  'input_spatial_with_cell_types': 'resources_test/task_ist_preprocessing/mouse_brain_combined/spatial_with_celltypes.h5ad',
+  'input': 'resources_test/task_ist_preprocessing/mouse_brain_combined/spatial_corrected.h5ad',
   'input_scrnaseq_reference': 'resources_test/task_ist_preprocessing/mouse_brain_combined/scrnaseq_reference.h5ad',
   'celltype_key': 'cell_type',
-  'output': 'spatial_corrected.h5ad',
+  'output': 'spatial_gene_efficiency_corrected.h5ad',
 }
 meta = {
   'name': 'gene_efficiency_correction',
 }
 ## VIASH END
 
-# Optional parameter check: For this specific correction method the par['input_sc'] is required
-assert par['input_scrnaseq_reference'] is not None, 'Single cell input is required for this expr correction method.'
-    
+# Optional parameter check: this correction method requires the scRNA-seq reference
+assert par['input_scrnaseq_reference'] is not None, 'Single cell input is required for this gene efficiency correction method.'
+
 # Read input
 print('Reading input files', flush=True)
-adata_sp = ad.read_h5ad(par['input_spatial_with_cell_types'])
+adata_sp = ad.read_h5ad(par['input'])
 adata_sc = ad.read_h5ad(par['input_scrnaseq_reference'])
-adata_sp.layers["normalized_uncorrected"] = adata_sp.layers["normalized"]
+# Preserve the pre-correction normalized layer. An upstream expression-correction stage
+# already sets it, so only initialise it here if it is missing (keeps it meaning "before
+# any correction").
+if "normalized_uncorrected" not in adata_sp.layers:
+    adata_sp.layers["normalized_uncorrected"] = adata_sp.layers["normalized"]
 adata_sp_reduced = adata_sp[:,adata_sc.var_names].copy()
 obs_sp = adata_sp.obs.copy()
 
 # Apply gene efficiency correction
-print('Annotating cell types', flush=True)
+print('Applying gene efficiency correction', flush=True)
 adata_sp_reduced = tx.preprocessing.gene_efficiency_correction(
     adata_sp_reduced, adata_sc, layer_key='normalized', ct_key=par['celltype_key']
 )
@@ -36,7 +40,7 @@
 print('Concatenating and reordering', flush=True)
 gene_order = adata_sp.var_names.tolist()
 gene_mask = ~adata_sp.var_names.isin(adata_sp_reduced.var_names)
-adata_sp = ad.concat([adata_sp[:,gene_mask], adata_sp_reduced], axis=1, join="outer") 
+adata_sp = ad.concat([adata_sp[:,gene_mask], adata_sp_reduced], axis=1, join="outer")
 adata_sp = adata_sp[:,gene_order]
 adata_sp.obs = obs_sp
 

src/methods_gene_efficiency_correction/no_correction/config.vsh.yaml

@@ -0,0 +1,31 @@
+__merge__: /src/api/comp_method_gene_efficiency_correction.yaml
+
+name: no_correction
+label: "No Correction"
+summary: "Dummy method that does not apply gene efficiency correction"
+description: >-
+  Dummy method that does not apply gene efficiency correction; passes the (expression-)corrected
+  input through unchanged.
+links:
+  documentation: "https://github.com/openproblems-bio/task_ist_preprocessing"
+  repository: "https://github.com/openproblems-bio/task_ist_preprocessing"
+references:
+  doi: "10.1101/2023.02.13.528102"
+
+resources:
+  - type: python_script
+    path: script.py
+
+engines:
+  - type: docker
+    image: openproblems/base_python:1
+    setup:
+      - type: python
+        pypi: [anndata]
+  - type: native
+
+runners:
+  - type: executable
+  - type: nextflow
+    directives:
+      label: [ midtime, lowcpu, lowmem ]

src/methods_gene_efficiency_correction/no_correction/script.py

@@ -0,0 +1,19 @@
+import anndata as ad
+
+## VIASH START
+par = {
+  'input': 'resources_test/task_ist_preprocessing/mouse_brain_combined/spatial_corrected.h5ad',
+  'output': 'spatial_gene_efficiency_corrected.h5ad',
+}
+## VIASH END
+
+# Read input
+print('Reading input files', flush=True)
+adata_sp = ad.read_h5ad(par['input'])
+# Preserve the pre-correction normalized layer only if an upstream stage hasn't set it.
+if "normalized_uncorrected" not in adata_sp.layers:
+    adata_sp.layers["normalized_uncorrected"] = adata_sp.layers["normalized"]
+
+# Write output
+print('Writing output', flush=True)
+adata_sp.write(par['output'])

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
@@ -78,11 +79,15 @@
 trans = sd.transformations.Sequence([trans_segm_to_global, trans_global_to_transcripts])
 boundaries = sd.transform(boundaries, trans, par['coordinate_system'])
 
-# Xenium boundary schema: one row per polygon vertex (cell_id int64, vertex_x/y float64).
+# Xenium boundary schema: one row per polygon vertex (vertex_x/y float64). cell_id is cast
+# to STRING (via int64, so labels render as clean "1"/"2") to match the transcripts.parquet
+# cell_id written below: segger joins transcript cell_id -> boundary cell_id to map each cell
+# to its polygon, and a str-vs-int mismatch makes every row miss -> all-nan entity_index ->
+# "Index has duplicate keys: [nan]" in setup_anndata. Real Xenium v2+ uses string ids in both.
 boundaries.index.name = "cell_id"
 boundaries_df = boundaries['geometry'].get_coordinates().rename(columns={'x': 'vertex_x', 'y': 'vertex_y'})
 boundaries_df = boundaries_df.reset_index()
-boundaries_df['cell_id'] = boundaries_df['cell_id'].astype(np.int64)
+boundaries_df['cell_id'] = boundaries_df['cell_id'].astype(np.int64).astype(str)
 boundaries_df['vertex_x'] = boundaries_df['vertex_x'].astype(np.float64)
 boundaries_df['vertex_y'] = boundaries_df['vertex_y'].astype(np.float64)
 boundaries_df = boundaries_df[['cell_id', 'vertex_x', 'vertex_y']]
@@ -154,6 +159,26 @@
 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   #
 ################