add magic with reverse norm order (#797)

Former-commit-id: 458fd8d

Author: scottgigante-immunai

openproblems/tasks/denoising/methods/__init__.py

@@ -7,3 +7,5 @@
 from .magic import knn_naive
 from .magic import magic
 from .magic import magic_approx
+from .magic import magic_approx_reverse_norm
+from .magic import magic_reverse_norm

openproblems/tasks/denoising/methods/magic.py

@@ -16,7 +16,7 @@
 )
 
 
-def _magic(adata, solver, normtype="sqrt", **kwargs):
+def _magic(adata, solver, normtype="sqrt", reverse_norm_order=False, **kwargs):
     from magic import MAGIC
 
     if normtype == "sqrt":
@@ -28,11 +28,19 @@ def _magic(adata, solver, normtype="sqrt", **kwargs):
     else:
         raise NotImplementedError
 
-    X, libsize = scprep.normalize.library_size_normalize(
-        adata.obsm["train"], rescale=1, return_library_size=True
-    )
+    X = adata.obsm["train"]
+    if reverse_norm_order:
+        # inexplicably, this sometimes performs better
+        X = scprep.utils.matrix_transform(X, norm_fn)
+        X, libsize = scprep.normalize.library_size_normalize(
+            X, rescale=1, return_library_size=True
+        )
+    else:
+        X, libsize = scprep.normalize.library_size_normalize(
+            X, rescale=1, return_library_size=True
+        )
+        X = scprep.utils.matrix_transform(X, norm_fn)
 
-    X = scprep.utils.matrix_transform(X, norm_fn)
     Y = MAGIC(solver=solver, **kwargs, verbose=False).fit_transform(
         X, genes="all_genes"
     )
@@ -52,13 +60,27 @@ def magic(adata, test=False):
     return _magic(adata, solver="exact", normtype="sqrt")
 
 
+@_magic_method(
+    method_name="MAGIC (reversed normalization)",
+)
+def magic_reverse_norm(adata, test=False):
+    return _magic(adata, solver="exact", normtype="sqrt", reverse_norm_order=True)
+
+
 @_magic_method(
     method_name="MAGIC (approximate)",
 )
 def magic_approx(adata, test=False):
     return _magic(adata, solver="approximate", normtype="sqrt")
 
 
+@_magic_method(
+    method_name="MAGIC (approximate, reversed normalization)",
+)
+def magic_approx_reverse_norm(adata, test=False):
+    return _magic(adata, solver="approximate", normtype="sqrt", reverse_norm_order=True)
+
+
 @method(
     method_name="KNN smoothing",
     paper_name="KNN Smoothing (baseline)",