Add SenKin CITE-seq method (LightGBM + GRU ensemble)

scripts/create_datasets/test_resources.sh

@@ -90,6 +90,24 @@ for name in bmmc_cite/normal bmmc_cite/swap bmmc_multiome/normal bmmc_multiome/s
       --output $DATASET_DIR/$name/models/simple_mlp/
   fi
 
+  # senkin_tmp is CITE-only
+  if [[ "$name" == bmmc_cite/normal ]]; then
+    echo "pre-train senkin_tmp on $name"
+    if up_to_date $DATASET_DIR/$name/models/senkin_tmp/model.pkl $STATE; then
+      echo "  already up to date, skipping"
+    else
+      mkdir -p $DATASET_DIR/$name/models/senkin_tmp/
+      viash run src/methods/senkin_tmp/senkin_tmp_train/config.vsh.yaml -- \
+        --input_train_mod1 $DATASET_DIR/$name/train_mod1.h5ad \
+        --input_train_mod2 $DATASET_DIR/$name/train_mod2.h5ad \
+        --input_test_mod1 $DATASET_DIR/$name/test_mod1.h5ad \
+        --lgbm_boost_rounds 50 \
+        --lgbm_early_stopping 10 \
+        --nn_epochs 2 \
+        --output $DATASET_DIR/$name/models/senkin_tmp/model.pkl
+    fi
+  fi
+
   echo "pre-train novel on $name"
   if up_to_date $DATASET_DIR/$name/models/novel/ $STATE; then
     echo "  already up to date, skipping"

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

@@ -0,0 +1,27 @@
+__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)"
+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.
+references:
+  doi:
+    - 10.64898/2026.02.24.707614
+links:
+  repository: https://github.com/lueckenlab/senkin-tmp-cite-pred
+info:
+  preferred_normalization: log_cp10k
+resources:
+  - path: main.nf
+    type: nextflow_script
+    entrypoint: run_wf
+dependencies:
+  - name: methods/senkin_tmp_train
+  - name: methods/senkin_tmp_predict
+runners:
+  - type: nextflow

src/methods/senkin_tmp/senkin_tmp/main.nf

@@ -0,0 +1,18 @@
+workflow run_wf {
+  take: input_ch
+  main:
+  output_ch = input_ch
+    | senkin_tmp_train.run(
+      fromState: ["input_train_mod1", "input_train_mod2", "input_test_mod1"],
+      toState: ["input_model": "output"]
+    )
+    | senkin_tmp_predict.run(
+      fromState: ["input_test_mod1", "input_train_mod2", "input_model"],
+      toState: ["output": "output"]
+    )
+    | map { tup ->
+      [tup[0], [output: tup[1].output]]
+    }
+
+  emit: output_ch
+}

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

@@ -0,0 +1,34 @@
+__merge__: ../../../api/comp_method_predict.yaml
+name: senkin_tmp_predict
+
+info:
+  test_setup:
+    with_model:
+      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: resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal/models/senkin_tmp/model.pkl
+resources:
+  - path: script.py
+    type: python_script
+test_resources:
+  - path: /resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal
+    dest: resources_test/task_predict_modality/openproblems_neurips2021/bmmc_cite/normal
+engines:
+  - type: docker
+    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
+      - type: python
+        packages:
+          - lightgbm>=4.0
+          - tensorflow>=2.12
+          - scikit-learn>=1.1
+          - mudata>=0.2
+          - muon>=0.1
+          - fast-array-utils
+runners:
+  - type: executable
+  - type: nextflow
+    directives:
+      label: [highmem, hightime, midcpu]

src/methods/senkin_tmp/senkin_tmp_predict/script.py

@@ -0,0 +1,41 @@
+import logging
+import pickle
+
+import anndata as ad
+import numpy as np
+from scipy.sparse import csc_matrix
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+## 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_model": "output_model.pkl",
+    "output": "output_pred.h5ad",
+}
+meta = {"name": "senkin_tmp"}
+## VIASH END
+
+logger.info("Reading input files...")
+adata_rna_test   = ad.read_h5ad(par["input_test_mod1"])
+adata_prot_train = ad.read_h5ad(par["input_train_mod2"])
+
+logger.info("Loading model bundle...")
+with open(par["input_model"], "rb") as f:
+    bundle = pickle.load(f)
+
+logger.info("Writing predictions...")
+adata_out = ad.AnnData(
+    layers={"normalized": csc_matrix(bundle["test_predictions"])},
+    obs=adata_rna_test.obs,
+    var=adata_prot_train.var,
+    uns={
+        "dataset_id": adata_rna_test.uns.get("dataset_id", bundle.get("dataset_id", "")),
+        "method_id": "senkin_tmp",
+    },
+)
+
+adata_out.write_h5ad(par["output"], compression="gzip")
+logger.info("Predictions saved to %s", par["output"])

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

@@ -0,0 +1,51 @@
+__merge__: ../../../api/comp_method_train.yaml
+name: senkin_tmp_train
+resources:
+  - path: script.py
+    type: python_script
+arguments:
+  - name: "--n_folds"
+    type: integer
+    default: 5
+    description: Number of cross-validation folds for LightGBM and neural network training.
+  - name: "--lgbm_boost_rounds"
+    type: integer
+    default: 10000
+    description: Maximum LightGBM boosting rounds (early stopping applies).
+    info:
+      test_default: 50
+  - name: "--lgbm_early_stopping"
+    type: integer
+    default: 100
+    description: LightGBM early stopping patience (rounds without improvement).
+    info:
+      test_default: 10
+  - name: "--nn_epochs"
+    type: integer
+    default: 100
+    description: Maximum neural network training epochs (early stopping applies).
+    info:
+      test_default: 2
+  - name: "--n_tsvd_components"
+    type: integer
+    default: 100
+    description: TSVD components for reducing LightGBM predictions before NN input.
+engines:
+  - type: docker
+    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
+      - type: python
+        packages:
+          - lightgbm>=4.0
+          - tensorflow>=2.12
+          - scikit-learn>=1.1
+          - mudata>=0.2
+          - muon>=0.1
+          - fast-array-utils
+runners:
+  - type: executable
+  - type: nextflow
+    directives:
+      label: [highmem, hightime, midcpu]