Merge pull request #15 from quadbio/tests/extend

Extend tests & update the README

Author: Marius1311

CHANGELOG.md

@@ -10,6 +10,9 @@ and this project adheres to [Semantic Versioning][].
 
 ## [Unreleased]
 
+### Added
+- Included tests for the `check` module, and more tests for the main classes.
+
 ## [v0.1.1]
 
 ### Changed

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2025, Marius Lange
+Copyright (c) 2025, QuaDBioLab
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -10,34 +10,46 @@
 [badge-pre-commit]: https://results.pre-commit.ci/badge/github/quadbio/cellmapper/main.svg
 [badge-pypi]: https://img.shields.io/pypi/v/cellmapper.svg
 
-k-NN-based mapping of cells across representations to tranfer labels, embeddings and expression values. Works for millions of cells, on CPU and GPU, across molecular modalities, between spatial and non-spatial data, for arbitrary query and reference datasets. Using `faiss` to compute k-NN graphs, CellMapper takes about 30 seconds to transfer cell type labels from 1.5M cells to 1.5M cells on a single RTX 4090 with 60 GB CPU memory.
+k-NN-based mapping of cells across representations to tranfer labels, embeddings and expression values. Works for millions of cells, on CPU and GPU, across molecular modalities, between spatial and non-spatial data, for arbitrary query and reference datasets. Using [faiss][] to compute k-NN graphs, CellMapper takes about 30 seconds to transfer cell type labels from 1.5M cells to 1.5M cells on a single RTX 4090 with 60 GB CPU memory.
 
-## Getting started
-
-Please refer to the [documentation][],
-in particular, the [API documentation][].
+Inspired by scanpy's [ingest][] and the [HNOCA-tools][] packages.
 
 ## Installation
 
 You need to have Python 3.10 or newer installed on your system.
 If you don't have Python installed, we recommend installing [uv][].
 
-There are several alternative options to install cellmapper:
+There are two alternative options to install ``cellmapper``:
 
-<!--
-1) Install the latest release of `cellmapper` from [PyPI][]:
+- **Install the latest release from [PyPI][]**:
 
-```bash
-pip install cellmapper
-```
--->
+  ```bash
+  pip install cellmapper
+  ```
+
+- **Install the latest development version**:
+
+  ```bash
+  pip install git+https://github.com/quadbio/cellmapper.git@main
+  ```
 
-1. Install the latest development version:
+## Getting started
+
+This package assumes that you have ``ref`` and ``query`` AnnData objects, with a joint embedding computed and stored in ``.obsm``. We explicilty do not compute this joint embedding, but there are plenty of method you can use to get such joint embeddings, e.g. [GimVI][] or [ENVI][] for spatial mapping, [GLUE][], [MIDAS][] and [MOFA+][] for modality translation, and [scVI][], [scANVI][] and [scArches][] for query-to-reference mapping - this is just a small selection!
+
+With a joint embedding in ``.obsm["X_joint"]`` at hand, the simplest way to use ``CellMapper`` is as follows:
+```Python
+from cellmapper import CellMapper
 
-```bash
-pip install git+https://github.com/quadbio/cellmapper.git@main
+cmap = CellMapper(ref, query).fit(
+    use_rep="X_joint", obs_keys="celltype", obsm_keys="X_umap", layer_key="X"
+    )
 ```
 
+This will transfer data from the reference to the query dataset, including celltype labels stored in ``ref.obs``, a UMAP embedding stored in ``ref.obsm``, and expression values stored in ``ref.X``.
+
+There are many ways to customize this, e.g. use different ways to compute k-NN graphs and to turn them into mapping matrices, and we implement a few methods to evaluate whether your k-NN transfer was sucessful.
+
 ## Release notes
 
 See the [changelog][].
@@ -59,3 +71,16 @@ Please cite this GitHub repo if you find CellMapper useful for your research.
 [coverage]: https://codecov.io/gh/quadbio/cellmapper
 [pre-commit]: https://results.pre-commit.ci/latest/github/quadbio/cellmapper/main
 [pypi]: https://pypi.org/project/cellmapper/
+[faiss]: https://github.com/facebookresearch/faiss
+
+[ingest]: https://scanpy.readthedocs.io/en/stable/generated/scanpy.tl.ingest.html
+[HNOCA-tools]: https://devsystemslab.github.io/HNOCA-tools/
+
+[GimVI]: https://docs.scvi-tools.org/en/stable/api/reference/scvi.external.GIMVI.html#
+[ENVI]: https://scenvi.readthedocs.io/en/latest/#
+[GLUE]: https://scglue.readthedocs.io/en/latest/
+[MIDAS]: https://scmidas.readthedocs.io/en/latest/
+[MOFA+]: https://muon.readthedocs.io/en/latest/omics/multi.html
+[scVI]: https://docs.scvi-tools.org/en/stable/api/reference/scvi.model.SCVI.html
+[scANVI]: https://docs.scvi-tools.org/en/stable/api/reference/scvi.model.SCANVI.html
+[scArches]: https://docs.scarches.org/en/latest/

src/cellmapper/check.py

@@ -79,4 +79,6 @@ def check_deps(*args) -> None:
         A list of dependencies to check
     """
     for item in args:
+        if item not in CHECKERS:
+            raise RuntimeError(f"Dependency '{item}' is not registered in CHECKERS.")
         CHECKERS[item].check()

src/cellmapper/knn.py

@@ -118,7 +118,9 @@ def knn_graph_connectivities(
             epsilon = kwargs.get("epsilon", 1e-8)
             connectivities = 1.0 / (self.distances + epsilon)
         else:
-            raise ValueError(f"Unknown kernel: {kernel}. Supported kernels are 'gaussian' and 'scarches'.")
+            raise ValueError(
+                f"Unknown kernel: {kernel}. Supported kernels are: 'gaussian', 'scarches', 'random', 'inverse_distance'."
+            )
         rowptr = np.arange(0, self.n_samples * self.n_neighbors + 1, self.n_neighbors)
         return csr_matrix((connectivities.ravel().astype(dtype), self.indices.ravel(), rowptr), shape=self.shape)
 

tests/test_cellmapper.py

@@ -1,6 +1,8 @@
 import numpy as np
 import pytest
 
+from cellmapper.cellmapper import CellMapper
+
 
 def assert_metrics_close(actual: dict, expected: dict, atol=1e-3):
     for key, exp in expected.items():
@@ -66,3 +68,48 @@ def test_compute_neighbors_joint_pca(self, cmap, joint_pca_key, n_pca_components
         assert joint_pca_key in cmap.query.obsm
         assert cmap.ref.obsm[joint_pca_key].shape[1] == n_pca_components
         assert cmap.query.obsm[joint_pca_key].shape[1] == n_pca_components
+
+    @pytest.mark.parametrize(
+        "obs_keys,obsm_keys,layer_key",
+        [
+            ("leiden", None, None),
+            (None, "X_pca", None),
+            (None, None, "X"),
+            ("leiden", "X_pca", None),
+            ("leiden", None, "X"),
+            (None, "X_pca", "X"),
+            ("leiden", "X_pca", "X"),
+        ],
+    )
+    def test_fit_various_combinations(self, cmap, obs_keys, obsm_keys, layer_key):
+        cmap.fit(obs_keys=obs_keys, obsm_keys=obsm_keys, layer_key=layer_key)
+        if obs_keys is not None:
+            keys = [obs_keys] if isinstance(obs_keys, str) else obs_keys
+            for key in keys:
+                assert f"{key}_pred" in cmap.query.obs
+        if obsm_keys is not None:
+            keys = [obsm_keys] if isinstance(obsm_keys, str) else obsm_keys
+            for key in keys:
+                assert f"{key}_pred" in cmap.query.obsm
+        if layer_key is not None:
+            assert cmap.query_imputed is not None
+            assert cmap.query_imputed.X.shape[0] == cmap.query.n_obs
+
+    def test_transfer_labels_self_mapping(self, query_ref_adata):
+        """Check mapping to self."""
+        _, ref = query_ref_adata
+        cm = CellMapper(ref, ref)
+        cm.fit(
+            knn_method="sklearn",
+            mapping_method="jaccard",
+            obs_keys="leiden",
+            use_rep="X_pca",
+            n_neighbors=1,
+            prediction_postfix="transfer",
+        )
+        assert "leiden_transfer" in ref.obs
+        assert len(ref.obs["leiden_transfer"]) == len(ref.obs["leiden"])
+        # Check that all predicted labels are valid categories
+        assert set(ref.obs["leiden_transfer"].cat.categories) <= set(ref.obs["leiden"].cat.categories)
+        # If mapping to self, labels should match
+        assert ref.obs["leiden_transfer"].equals(ref.obs["leiden"])

tests/test_check.py

@@ -0,0 +1,34 @@
+import packaging
+import pytest
+
+from cellmapper import check
+from cellmapper.check import Checker, check_deps
+
+
+class TestCheck:
+    def test_checker_available_module(self):
+        # Should not raise for a real installable package
+        Checker("packaging").check()
+
+    def test_checker_missing_module(self):
+        # Should raise RuntimeError for a missing module
+        with pytest.raises(RuntimeError):
+            Checker("not_a_real_module").check()
+
+    def test_checker_version_requirement(self):
+        # Should raise if vmin is higher than installed version
+        installed_version = packaging.version.parse(packaging.__version__)
+        higher_version = str(installed_version.major + 1) + ".0.0"
+        with pytest.raises(RuntimeError):
+            Checker("packaging", vmin=higher_version).check()
+
+    def test_check_deps_missing(self):
+        # Should raise for a missing dependency (not registered in CHECKERS)
+        with pytest.raises(RuntimeError):
+            check_deps("not_a_real_module")
+
+    def test_check_deps_available(self):
+        # Should not raise for a real installable package
+        check.CHECKERS["packaging"] = Checker("packaging")
+        check_deps("packaging")
+        del check.CHECKERS["packaging"]

tests/test_neighbors.py

@@ -1,4 +1,5 @@
 import numpy as np
+import pytest
 
 from cellmapper.knn import Neighbors
 
@@ -11,17 +12,27 @@ def assert_adjacency_equal(neigh1, neigh2, attrs=("xx", "yy", "xy", "yx")):
 
 
 class TestNeighbors:
-    def test_neighbors_sklearn_vs_pynndescent(self, small_data):
+    @pytest.mark.parametrize("only_yx", [False, True])
+    def test_neighbors_sklearn_vs_pynndescent(self, small_data, only_yx):
         x, y = small_data
         n_neighbors = 3
         # sklearn
         neigh_skl = Neighbors(x, y)
-        neigh_skl.compute_neighbors(n_neighbors=n_neighbors, method="sklearn")
+        neigh_skl.compute_neighbors(n_neighbors=n_neighbors, method="sklearn", only_yx=only_yx)
         # pynndescent
         neigh_pynn = Neighbors(x, y)
-        neigh_pynn.compute_neighbors(n_neighbors=n_neighbors, method="pynndescent")
-        # Compare adjacency matrices
-        assert_adjacency_equal(neigh_skl, neigh_pynn)
+        neigh_pynn.compute_neighbors(n_neighbors=n_neighbors, method="pynndescent", only_yx=only_yx)
+        if only_yx:
+            with pytest.raises(ValueError):
+                neigh_skl.get_adjacency_matrices()
+            with pytest.raises(ValueError):
+                neigh_pynn.get_adjacency_matrices()
+        else:
+            assert_adjacency_equal(neigh_skl, neigh_pynn)
+        # Always compare connectivities (yx)
+        conn_skl = neigh_skl.yx.knn_graph_connectivities()
+        conn_pynn = neigh_pynn.yx.knn_graph_connectivities()
+        assert np.allclose(conn_skl.toarray(), conn_pynn.toarray(), atol=1e-6)
 
     def test_neighbors_repr(self, small_data):
         x, y = small_data