fixes to make MultiVI run

.github/workflows/integration01-ci.yml

@@ -36,11 +36,13 @@ jobs:
           channel-priority: strict
           activate-environment: pipeline_env
           environment-file: pipeline_env.yaml
-
+# important: this patch is only to test if multivi integration works
+# issues are not related to panpipes https://discourse.scverse.org/t/error-when-training-model-on-m3-max-mps/1896/2
+# https://discourse.scverse.org/t/macbook-m1-m2-mps-acceleration-with-scvi/2075/4
       - name: Install Panpipes
         shell: bash -el {0}
         run: |
-          pip install -e .
+          pip install '.[multivipatch]' 
           conda list
 
       - name: Conda info

panpipes/python_scripts/batch_correct_multivi.py

@@ -78,9 +78,12 @@
 del mdata
 
 if check_for_bool(params["multimodal"]["MultiVI"]["lowmem"]):
-    L.info("subsetting atac to top 25k HVF")
+    if 'hvg' in atac.uns.keys():
+        L.info("subsetting atac to top HVF")
+        atac = atac[:, atac.var.highly_variable].copy()
+    L.info("calculating and subsetting atac to top 25k HVF")
     sc.pp.highly_variable_genes(atac, n_top_genes=25000)
-    atac = atac[:, atac.var.highly_variable]
+    atac = atac[:, atac.var.highly_variable].copy()
 
 
 
@@ -125,6 +128,15 @@
 L.info("concatenating modalities to comply with multiVI")
 # adata_paired = ad.concat([rna, atac], join="outer")
 # adata_paired.var = pd.concat([rna.var,atac.var])
+if rna.is_view:
+    L.info("RNA is view")
+    atac = rna.copy()
+if atac.is_view:
+    L.info("ATAC is view")
+    atac = atac.copy()
+
+L.info(atac.is_view)
+
 
 adata_paired = ad.concat([rna.T, atac.T]).T
 
@@ -222,14 +234,14 @@
     multivi_training_args={}
 else:
     multivi_training_args =  {k: v for k, v in params["multimodal"]['MultiVI']['training_args'].items() if v is not None}
-
+L.info("multivi training args")
 print(multivi_training_args)
 
 if params["multimodal"]['MultiVI']['training_plan'] is None:
     multivi_training_plan = {}
 else:
     multivi_training_plan =  {k: v for k, v in params["multimodal"]['MultiVI']['training_plan'].items() if v is not None}
-
+L.info("multivi training plan")
 print(multivi_training_plan)
 
 mvi.view_anndata_setup()

pyproject.toml

@@ -41,7 +41,7 @@ dependencies = [
     "scikit-misc",
     "scirpy",
     "scrublet",
-    "scvi-tools>=1.1.1",
+    "scvi-tools", #>=1.1.1
     "sqlalchemy",
     ]
 
@@ -61,6 +61,11 @@ spatial = [
     "cell2location",
     "tangram-sc"
 ]
+
+multivipatch = [
+    "scvi-tools<=0.20.3",
+    "requests"
+]
 [project.scripts]
 panpipes = "panpipes:entry.main"
 

tests/integration_1/pipeline.yml

@@ -211,7 +211,7 @@ multimodal:
   # specify one at a time and run the pipeline in different folders i.e. integration_by_sample, 
   # integration_by_tissue ...
   column_continuous: 
-  column_categorical: sample_id 
+  column_categorical: dataset
   # extra params:
   totalvi:
     # this is a minimal set of parameters that will be expected
@@ -269,8 +269,8 @@ multimodal:
       batch_size : 128
       #float (default: 0.001)
       weight_decay : 0.001 
-      #float (default: 1e-08)
-      eps : 1e-08 
+      #float (default: 1.0e-08)
+      eps : 1.0e-08 
       #bool (default: True)
       early_stopping : True 
       #bool (default: True)