add transformer

src/transformers/transform/config.vsh.yaml

@@ -0,0 +1,20 @@
+__merge__: /src/api/comp_transformer.yaml
+name: transform
+label: Transform outputs
+summary: Transform batch integration outputs where necessary
+description: |
+  Transform corrected feature output to an embedding, and an embedding to a graph output.
+resources:
+  - type: python_script
+    path: script.py
+engines:
+  - type: docker
+    image: openproblems/base_python:1.0.0
+    setup:
+      - type: python
+        pypi: scanpy
+runners:
+  - type: executable
+  - type: nextflow
+    directives: 
+      label: [midtime, midmem, lowcpu]

src/transformers/transform/script.py

@@ -0,0 +1,37 @@
+import anndata as ad
+import scanpy as sc
+
+## VIASH START
+par = {
+    'input': 'resources_test/task_batch_integration/cxg_mouse_pancreas_atlas/integrated.h5ad',
+    'ouput': 'output.h5ad'
+}
+## VIASH END
+
+print('Read input', flush=True)
+adata = ad.read_h5ad(par['input'])
+
+assert 'corrected_counts' in adata.layers.keys() or \
+    'X_emb' in adata.obsm.keys() or \
+    'neighbors' in adata.uns.keys(), \
+    "Input data is missing necessary information. " \
+    "Expected 'corrected_counts' in layers, " \
+    "'X_emb' in obsm, or 'neighbors' in uns.\n" \
+    f"Found: {adata}"
+
+if 'X_emb' not in adata.obsm:
+    print('Run PCA...', flush=True)
+    adata.obsm['X_emb'] = sc.pp.pca(
+        adata.layers['corrected_counts'],
+        n_comps=50,
+        use_highly_variable=False,
+        svd_solver='arpack',
+        return_info=False
+    )
+
+if 'neighbors' not in adata.uns:
+    print('Run kNN...', flush=True)
+    sc.pp.neighbors(adata, use_rep='X_emb')
+
+print("Store outputs", flush=True)
+adata.write_h5ad(par['output'], compression='gzip')