Fix/senkin match original

src/methods/senkin_tmp/senkin_tmp/config.vsh.yaml

@@ -1,14 +1,17 @@
 __merge__: ../../../api/comp_method.yaml
 name: senkin_tmp
 label: "senkin & tmp"
-summary: "LightGBM + bidirectional GRU ensemble for CITE-seq protein prediction (OpenProblems 2022 2nd place)"
+summary: "LightGBM + bidirectional GRU ensemble for CITE-seq protein prediction (best CITE-seq model of the Open Problems 2022 competition)"
 description: |
-  Two-stage method from the OpenProblems NeurIPS 2021 competition. Stage 1 trains four
-  LightGBM models on different RNA feature representations (log-normalized, CLR-TSVD,
-  custom sqrt-normalized, and raw counts). Stage 2 refines predictions with two neural
-  network architectures: a bidirectional GRU with cosine-similarity loss and a dense
-  bidirectional GRU with MSE loss. Final predictions are a weighted blend (55% cosine,
-  45% MSE) of per-fold averaged outputs.
+  Two-stage method of senkin13 from the Open Problems NeurIPS 2022 multimodal single-cell
+  integration competition (2nd place overall, best CITE-seq submission). Stage 1 trains four
+  LightGBM models on different RNA feature representations (log-normalized counts; CLR-TSVD
+  together with correlated genes and TSVD/PCA of a custom sqrt normalization with per-batch
+  median correction; raw counts; raw counts predicting raw protein counts). Stage 2 refines
+  the TSVD-reduced LightGBM predictions, together with the feature blocks (each z-scored per
+  cell), with two neural networks: a bidirectional GRU with cosine-similarity loss and a dense
+  bidirectional GRU with MSE loss. Final predictions are a weighted blend (55% cosine, 45% MSE)
+  of per-fold averaged outputs, rescaled to the target scale with a global affine transform.
 references:
   doi:
     - 10.64898/2026.02.24.707614

src/methods/senkin_tmp/senkin_tmp_predict/config.vsh.yaml

@@ -18,7 +18,7 @@ engines:
     image: openproblems/base_pytorch_nvidia:1
     setup:
       - type: docker
-        run: pip install --no-cache-dir --no-deps git+https://github.com/lueckenlab/senkin-tmp-cite-pred.git
+        run: pip install --no-cache-dir --no-deps git+https://github.com/lueckenlab/senkin-tmp-cite-pred.git@a830343a148c087dd95f60d5c7e14529aaba2425
       - type: python
         packages:
           - lightgbm>=4.0
@@ -31,4 +31,4 @@ runners:
   - type: executable
   - type: nextflow
     directives:
-      label: [highmem, hightime, midcpu]
+      label: [lowmem, lowtime, lowcpu]

src/methods/senkin_tmp/senkin_tmp_predict/script.py

@@ -10,8 +10,8 @@
 
 ## VIASH START
 par = {
-    "input_test_mod1": "resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/swap/test_mod1.h5ad",
-    "input_train_mod2": "resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/swap/train_mod2.h5ad",
+    "input_test_mod1": "resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal/test_mod1.h5ad",
+    "input_train_mod2": "resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal/train_mod2.h5ad",
     "input_model": "output_model.pkl",
     "output": "output_pred.h5ad",
 }
@@ -26,9 +26,15 @@
 with open(par["input_model"], "rb") as f:
     bundle = pickle.load(f)
 
+predictions = bundle["test_predictions"]
+if "test_obs_names" in bundle:
+    # The train step predicted the test cells in the order of input_test_mod1; make sure nothing changed
+    assert (np.asarray(bundle["test_obs_names"]) == adata_rna_test.obs_names.values).all(), "test cells do not match the trained model"
+assert predictions.shape == (adata_rna_test.n_obs, adata_prot_train.n_vars)
+
 logger.info("Writing predictions...")
 adata_out = ad.AnnData(
-    layers={"normalized": csc_matrix(bundle["test_predictions"])},
+    layers={"normalized": csc_matrix(predictions)},
     obs=adata_rna_test.obs,
     var=adata_prot_train.var,
     uns={

src/methods/senkin_tmp/senkin_tmp_train/config.vsh.yaml

@@ -1,8 +1,18 @@
 __merge__: ../../../api/comp_method_train.yaml
 name: senkin_tmp_train
+info:
+  # The api default test inputs are bmmc_cite/swap (ADT->GEX). senkin predicts protein
+  # (ADT) from RNA (GEX), so point the component test at the 'normal' (GEX->ADT)
+  # direction, otherwise the new modality guard correctly exits non-applicable.
+  test_setup:
+    normal_direction:
+      input_train_mod1: resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal/train_mod1.h5ad
+      input_train_mod2: resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal/train_mod2.h5ad
+      input_test_mod1: resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal/test_mod1.h5ad
 resources:
   - path: script.py
     type: python_script
+  - path: /src/utils/exit_codes.py
 arguments:
   - name: "--n_folds"
     type: integer
@@ -56,9 +66,15 @@ engines:
         github:
           - openproblems-bio/core#subdirectory=packages/python/openproblems
       - type: docker
-        run: pip install --no-cache-dir --no-deps git+https://github.com/lueckenlab/senkin-tmp-cite-pred.git
+        run: pip install --no-cache-dir --no-deps git+https://github.com/lueckenlab/senkin-tmp-cite-pred.git@a830343a148c087dd95f60d5c7e14529aaba2425
 runners:
   - type: executable
   - type: nextflow
     directives:
       label: [highmem, hightime, midcpu, gpu]
+# Override the api default (bmmc_cite/swap = ADT->GEX). senkin only supports GEX->ADT.
+test_resources:
+  - type: python_script
+    path: /common/component_tests/run_and_check_output.py
+  - path: /resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal
+    dest: resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal