Merge pull request #158 from YeoLab/spatial-multimap

multimapping shapes

.gitignore

@@ -23,7 +23,6 @@ lib/
 lib64/
 parts/
 sdist/
-var/
 wheels/
 pip-wheel-metadata/
 share/python-wheels/

bento/geometry/_ops.py

@@ -20,6 +20,7 @@ def overlay(
     name: str,
     how: str = "intersection",
     make_valid: bool = True,
+    instance_map_type: str = "1to1",
 ):
     """Overlay two shape elements in a SpatialData object and store the result as a new shape element.
 
@@ -37,6 +38,8 @@ def overlay(
         If True, correct invalid geometries with GeoPandas, by default True
     instance_key : str
         Name of the shape element to use as the instance for indexing, by default "cell_boundaries". If None, no indexing is performed.
+    instance_map_type : str, optional
+        Type of instance mapping to use. Options are "1to1", "1tomany", by default "1to1".
 
     Returns
     -------
@@ -46,7 +49,8 @@ def overlay(
     shape1 = sdata[s1]
     shape2 = sdata[s2]
 
-    new_shape = shape1.overlay(shape2, how=how, make_valid=make_valid)
+    new_shape = shape1.overlay(shape2, how=how, make_valid=make_valid)[["geometry"]]
+    new_shape.index = new_shape.index.astype(str)
     new_shape.attrs = {}
 
     transform = shape1.attrs
@@ -58,6 +62,7 @@ def overlay(
         shape_keys=[name],
         instance_key=get_instance_key(sdata),
         feature_key=get_feature_key(sdata),
+        instance_map_type=instance_map_type,
     )
 
 

bento/io/_index.py

@@ -1,9 +1,9 @@
-from typing import List
+from typing import List, Union
 
 import pandas as pd
 import geopandas as gpd
 from spatialdata._core.spatialdata import SpatialData
-
+from spatialdata.models import ShapesModel
 from .._utils import (
     get_points,
     set_points_metadata,
@@ -68,7 +68,12 @@ def _sjoin_points(
     return sdata
 
 
-def _sjoin_shapes(sdata: SpatialData, instance_key: str, shape_keys: List[str]):
+def _sjoin_shapes(
+    sdata: SpatialData,
+    instance_key: str,
+    shape_keys: List[str],
+    instance_map_type: Union[str, dict],
+):
     """Adds polygon indexes to sdata.shapes[instance_key][shape_key] for point feature analysis.
 
     Parameters
@@ -102,17 +107,36 @@ def _sjoin_shapes(sdata: SpatialData, instance_key: str, shape_keys: List[str]):
 
     # sjoin shapes to instance_key shape
     for shape_key in shape_keys:
-        child_shape = sdata.shapes[shape_key]
+        child_shape = sdata.shapes[shape_key].copy()
+        child_attrs = child_shape.attrs
         # Hack for polygons that are 99% contained in parent shape or have shared boundaries
         child_shape = gpd.GeoDataFrame(geometry=child_shape.buffer(-10e-6))
 
         # Map child shape index to parent shape and process the result
+
+        if instance_map_type == "1tomany":
+            child_shape = (
+                child_shape.sjoin(
+                    parent_shape.reset_index(drop=True),
+                    how="left",
+                    predicate="covered_by",
+                )
+                .dissolve(by="index_right", observed=True, dropna=False)
+                .reset_index(drop=True)[["geometry"]]
+            )
+            child_shape.index = child_shape.index.astype(str)
+            child_shape = ShapesModel.parse(child_shape)
+            child_shape.attrs = child_attrs
+            sdata.shapes[shape_key] = child_shape
+
         parent_shape = (
             parent_shape.sjoin(child_shape, how="left", predicate="covers")
-            .reset_index()
-            .drop_duplicates(subset="index", keep="last")
+            .reset_index()  # ignore any user defined index name
+            .drop_duplicates(
+                subset="index", keep="last"
+            )  # Remove multiple child shapes mapped to same parent shape
             .set_index("index")
-            .assign(
+            .assign(  # can this just be fillna on index_right?
                 index_right=lambda df: df.loc[
                     ~df["index_right"].duplicated(keep="first"), "index_right"
                 ]
@@ -121,14 +145,20 @@ def _sjoin_shapes(sdata: SpatialData, instance_key: str, shape_keys: List[str]):
             )
             .rename(columns={"index_right": shape_key})
         )
+
+        if (
+            parent_shape[shape_key].dtype == "category"
+            and "" not in parent_shape[shape_key].cat.categories
+        ):
+            parent_shape[shape_key] = parent_shape[shape_key].cat.add_categories([""])
         parent_shape[shape_key] = parent_shape[shape_key].fillna("")
 
         # Save shape index as column in instance_key shape
         set_shape_metadata(
             sdata, shape_key=instance_key, metadata=parent_shape[shape_key]
         )
 
-        # Add instance_key shape index to shape
+        # Add instance_key shape index to child shape
         instance_index = (
             parent_shape.drop_duplicates(subset=shape_key)
             .reset_index()

bento/io/_io.py

@@ -1,7 +1,8 @@
 import warnings
-from typing import List
+from typing import List, Union
 
 import emoji
+import spatialdata as sd
 from anndata.utils import make_index_unique
 from spatialdata import SpatialData
 from spatialdata.models import TableModel
@@ -19,6 +20,7 @@ def prep(
     feature_key: str = "feature_name",
     instance_key: str = "cell_boundaries",
     shape_keys: List[str] = ["cell_boundaries", "nucleus_boundaries"],
+    instance_map_type: Union[dict, str] = "1to1",
 ) -> SpatialData:
     """Computes spatial indices for elements in SpatialData to enable usage of bento-tools.
 
@@ -36,6 +38,11 @@ def prep(
         Key for the shape that will be used as the instance for all indexing. Usually the cell shape.
     shape_keys : str, list
         List of shape names to index points to
+    instance_map_type : str, dict
+        Type of mapping to use for the instance shape. If "1to1", each instance shape will be mapped to a single shape at most.
+        If "1tomany", each instance shape will be mapped to one or more shapes;
+        multiple shapes mapped to the same instance shape will be merged into a single MultiPolygon.
+        Use a dict to specify different mapping types for each shape.
 
     Returns
     -------
@@ -51,6 +58,24 @@ def prep(
             shape_gdf[shape_key] = shape_gdf["geometry"]
         shape_gdf.index = make_index_unique(shape_gdf.index.astype(str))
 
+    transform = {
+        "global": sd.transformations.get_transformation(sdata.points[points_key])
+    }
+    if "global" in sdata.points[points_key].attrs["transform"]:
+        # Force points to 2D for Xenium data
+        if isinstance(transform["global"], sd.transformations.Scale):
+            transform = {
+                "global": sd.transformations.Scale(
+                    scale=transform.to_scale_vector(["x", "y"]), axes=["x", "y"]
+                )
+            }
+    sdata.points[points_key] = sd.models.PointsModel.parse(
+        sdata.points[points_key].compute().reset_index(drop=True),
+        coordinates={"x": "x", "y": "y"},
+        feature_key=feature_key,
+        transformations=transform,
+    )
+
     # sindex points and sjoin shapes if they have not been indexed or joined
     point_sjoin = []
     shape_sjoin = []
@@ -72,6 +97,19 @@ def prep(
     sdata.points[points_key].attrs["spatialdata_attrs"]["instance_key"] = instance_key
 
     pbar = tqdm(total=3)
+    if len(shape_sjoin) > 0:
+        pbar.set_description(
+            "Mapping shapes"
+        )  # Map shapes must happen first; manyto1 mapping resets shape index
+        sdata = _sjoin_shapes(
+            sdata=sdata,
+            instance_key=instance_key,
+            shape_keys=shape_sjoin,
+            instance_map_type=instance_map_type,
+        )
+
+    pbar.update()
+
     if len(point_sjoin) > 0:
         pbar.set_description("Mapping points")
         sdata = _sjoin_points(
@@ -82,14 +120,6 @@ def prep(
 
     pbar.update()
 
-    if len(shape_sjoin) > 0:
-        pbar.set_description("Mapping shapes")
-        sdata = _sjoin_shapes(
-            sdata=sdata, instance_key=instance_key, shape_keys=shape_sjoin
-        )
-
-    pbar.update()
-
     # Only keep points within instance_key shape
     _sync_points(sdata, points_key)
 

bento/tools/_shape_features.py

@@ -165,6 +165,7 @@ def density(sdata: SpatialData, shape_key: str, recompute: bool = False):
         .query(f"{shape_key} != 'None'")[shape_key]
         .value_counts()
         .compute()
+        .reindex_like(sdata.shapes[shape_key])
     )
     area(sdata, shape_key)
 

pyproject.toml

@@ -24,7 +24,7 @@ dependencies = [
     "upsetplot>=0.9.0",
     "xgboost>=2.0.3",
     "statsmodels>=0.14.1",
-    "scikit-learn>=1.4.2",
+    "scikit-learn<1.6.0",
     "ipywidgets>=8.1.5",
 ]
 license = "BSD-2-Clause"