napistu 0.2.5.dev7__py3-none-any.whl → 0.3.1.dev1__py3-none-any.whl

tests/test_scverse_loading.py

@@ -0,0 +1,778 @@
+import pytest
+
+import anndata
+import mudata
+import numpy as np
+import pandas as pd
+from scipy import sparse
+
+from napistu.scverse import loading
+from napistu.scverse.constants import ADATA, SCVERSE_DEFS
+
+
+@pytest.fixture
+def minimal_adata():
+    """Create a minimal AnnData object for testing."""
+    # Create random data
+    n_obs, n_vars = 10, 5
+    X = np.random.randn(n_obs, n_vars)
+
+    # Create observation and variable annotations
+    obs = pd.DataFrame(
+        {"cell_type": ["type_" + str(i) for i in range(n_obs)]},
+        index=["cell_" + str(i) for i in range(n_obs)],
+    )
+    var = pd.DataFrame(
+        {
+            "gene_name": ["gene_" + str(i) for i in range(n_vars)],
+            "ensembl_transcript": [
+                f"ENST{i:011d}" for i in range(n_vars)
+            ],  # Add ensembl_transcript
+        },
+        index=["gene_" + str(i) for i in range(n_vars)],
+    )
+
+    # Create AnnData object
+    adata = anndata.AnnData(X=X, obs=obs, var=var)
+
+    # Add multiple layers to test table_name specification
+    adata.layers["counts"] = np.random.randint(0, 100, size=(n_obs, n_vars))
+    adata.layers["normalized"] = np.random.randn(n_obs, n_vars)
+
+    # Add varm matrix (n_vars × n_features)
+    n_features = 3
+    varm_array = np.random.randn(n_vars, n_features)
+    adata.varm["gene_scores"] = varm_array
+    # Store column names separately since varm is a numpy array
+    adata.uns["gene_scores_features"] = ["score1", "score2", "score3"]
+
+    # Add variable pairwise matrices (varp)
+    # Dense correlation matrix (n_vars × n_vars)
+    correlations = np.random.rand(n_vars, n_vars)
+    # Make it symmetric for a correlation matrix
+    correlations = (correlations + correlations.T) / 2
+    adata.varp["correlations"] = correlations
+
+    # Sparse adjacency matrix
+    adjacency = sparse.random(n_vars, n_vars, density=0.2)
+    # Make it symmetric
+    adjacency = (adjacency + adjacency.T) / 2
+    adata.varp["adjacency"] = adjacency
+
+    return adata
+
+
+def test_load_raw_table_success(minimal_adata):
+    """Test successful loading of various table types."""
+    # Test identity table (X)
+    x_result = loading._load_raw_table(minimal_adata, "X")
+    assert isinstance(x_result, np.ndarray)
+    assert x_result.shape == minimal_adata.X.shape
+
+    # Test dict-like table with name
+    layer_result = loading._load_raw_table(minimal_adata, "layers", "counts")
+    assert isinstance(layer_result, np.ndarray)
+    assert layer_result.shape == minimal_adata.layers["counts"].shape
+
+
+def test_load_raw_table_errors(minimal_adata):
+    """Test error cases for loading tables."""
+    # Test invalid table type
+    with pytest.raises(ValueError, match="is not a valid AnnData attribute"):
+        loading._load_raw_table(minimal_adata, "invalid_type")
+
+    # Test missing table name when required
+    with pytest.raises(
+        ValueError, match="Multiple tables found.*and table_name is not specified"
+    ):
+        loading._load_raw_table(minimal_adata, "layers")
+
+
+def test_get_table_from_dict_attr_success(minimal_adata):
+    """Test successful retrieval from dict-like attributes."""
+    # Test getting specific table
+    result = loading._get_table_from_dict_attr(minimal_adata, "varm", "gene_scores")
+    assert isinstance(result, np.ndarray)
+    assert result.shape == (minimal_adata.n_vars, 3)
+
+
+def test_get_table_from_dict_attr_errors(minimal_adata):
+    """Test error cases for dict-like attribute access."""
+    # Test missing table name with multiple tables
+    with pytest.raises(
+        ValueError, match="Multiple tables found.*and table_name is not specified"
+    ):
+        loading._get_table_from_dict_attr(minimal_adata, "layers")
+
+    # Test nonexistent table name
+    with pytest.raises(ValueError, match="table_name 'nonexistent' not found"):
+        loading._get_table_from_dict_attr(minimal_adata, "layers", "nonexistent")
+
+
+def test_select_results_attrs_success(minimal_adata):
+    """Test successful selection of results attributes."""
+    # Test numpy array selection - shape should be (n_obs x n_vars)
+    array = np.random.randn(
+        minimal_adata.n_obs, minimal_adata.n_vars
+    )  # 10x5 to match minimal_adata
+    array_results_attrs = minimal_adata.obs.index[:3].tolist()
+    array_result = loading._select_results_attrs(
+        minimal_adata, array, "X", array_results_attrs
+    )
+    assert isinstance(array_result, pd.DataFrame)
+    assert array_result.shape[0] == minimal_adata.var.shape[0]
+    assert len(array_result.columns) == len(array_results_attrs)
+    # Check orientation - vars should be rows
+    assert list(array_result.index) == minimal_adata.var.index.tolist()
+
+    # Test varm selection
+    varm_results_attrs = ["score1", "score2"]
+    varm_features = minimal_adata.uns["gene_scores_features"]
+    # Get column indices for the requested features
+    varm_col_indices = [varm_features.index(attr) for attr in varm_results_attrs]
+    varm_result = loading._select_results_attrs(
+        minimal_adata,
+        minimal_adata.varm["gene_scores"],
+        "varm",
+        varm_results_attrs,
+        table_colnames=varm_features,
+    )
+    assert isinstance(varm_result, pd.DataFrame)
+    assert varm_result.shape == (
+        minimal_adata.n_vars,
+        2,
+    )  # All vars x selected features
+    assert list(varm_result.columns) == varm_results_attrs
+    assert list(varm_result.index) == minimal_adata.var.index.tolist()
+    # Check values match original
+    np.testing.assert_array_equal(
+        varm_result.values, minimal_adata.varm["gene_scores"][:, varm_col_indices]
+    )
+
+    # Test varp selection with dense matrix
+    varp_results_attrs = minimal_adata.var.index[:2].tolist()  # Select first two genes
+    varp_result = loading._select_results_attrs(
+        minimal_adata, minimal_adata.varp["correlations"], "varp", varp_results_attrs
+    )
+    assert isinstance(varp_result, pd.DataFrame)
+    assert varp_result.shape == (minimal_adata.n_vars, 2)  # All vars x selected genes
+    assert list(varp_result.columns) == varp_results_attrs
+    assert list(varp_result.index) == minimal_adata.var.index.tolist()
+    # Check values match original
+    np.testing.assert_array_equal(
+        varp_result.values,
+        minimal_adata.varp["correlations"][:, :2],  # First two columns
+    )
+
+    # Test varp selection with sparse matrix
+    sparse_result = loading._select_results_attrs(
+        minimal_adata, minimal_adata.varp["adjacency"], "varp", varp_results_attrs
+    )
+    assert isinstance(sparse_result, pd.DataFrame)
+    assert sparse_result.shape == (minimal_adata.n_vars, 2)
+    assert list(sparse_result.columns) == varp_results_attrs
+    assert list(sparse_result.index) == minimal_adata.var.index.tolist()
+
+    # Test full table selection (results_attrs=None)
+    full_varm_result = loading._select_results_attrs(
+        minimal_adata,
+        minimal_adata.varm["gene_scores"],
+        "varm",
+        table_colnames=minimal_adata.uns["gene_scores_features"],
+    )
+    assert isinstance(full_varm_result, pd.DataFrame)
+    assert full_varm_result.shape == (minimal_adata.n_vars, 3)
+    assert list(full_varm_result.columns) == minimal_adata.uns["gene_scores_features"]
+    assert list(full_varm_result.index) == minimal_adata.var.index.tolist()
+
+
+def test_select_results_attrs_errors(minimal_adata):
+    """Test error cases for results attribute selection."""
+    # Test invalid results attributes - shape should match minimal_adata
+    array = np.random.randn(minimal_adata.n_obs, minimal_adata.n_vars)
+    with pytest.raises(
+        ValueError, match="The following results attributes were not found"
+    ):
+        loading._select_results_attrs(minimal_adata, array, "X", ["nonexistent_attr"])
+
+    # Test invalid gene names for varp
+    with pytest.raises(
+        ValueError, match="The following results attributes were not found"
+    ):
+        loading._select_results_attrs(
+            minimal_adata,
+            minimal_adata.varp["correlations"],
+            "varp",
+            results_attrs=["nonexistent_gene"],
+        )
+
+    # Test missing table_colnames for varm
+    with pytest.raises(ValueError, match="table_colnames is required for varm tables"):
+        loading._select_results_attrs(
+            minimal_adata, minimal_adata.varm["gene_scores"], "varm", ["score1"]
+        )
+
+    # Test DataFrame for array-type table
+    with pytest.raises(ValueError, match="must be a numpy array, not a DataFrame"):
+        loading._select_results_attrs(
+            minimal_adata,
+            pd.DataFrame(minimal_adata.varm["gene_scores"]),
+            "varm",
+            ["score1"],
+            table_colnames=minimal_adata.uns["gene_scores_features"],
+        )
+
+
+def test_create_results_df(minimal_adata):
+    """Test DataFrame creation from different AnnData table types."""
+    # Test varm table
+    varm_attrs = ["score1", "score2"]
+    varm_features = minimal_adata.uns["gene_scores_features"]
+    # Get column indices for the requested features
+    varm_col_indices = [varm_features.index(attr) for attr in varm_attrs]
+    varm_array = minimal_adata.varm["gene_scores"][:, varm_col_indices]
+    varm_result = loading._create_results_df(
+        array=varm_array,
+        attrs=varm_attrs,
+        var_index=minimal_adata.var.index,
+        table_type=ADATA.VARM,
+    )
+    assert varm_result.shape == (minimal_adata.n_vars, len(varm_attrs))
+    pd.testing.assert_index_equal(varm_result.index, minimal_adata.var.index)
+    pd.testing.assert_index_equal(varm_result.columns, pd.Index(varm_attrs))
+    np.testing.assert_array_equal(varm_result.values, varm_array)
+
+    # Test varp table with dense correlations
+    varp_attrs = minimal_adata.var.index[:2].tolist()  # First two genes
+    varp_array = minimal_adata.varp["correlations"][:, :2]
+    varp_result = loading._create_results_df(
+        array=varp_array,
+        attrs=varp_attrs,
+        var_index=minimal_adata.var.index,
+        table_type=ADATA.VARP,
+    )
+    assert varp_result.shape == (minimal_adata.n_vars, len(varp_attrs))
+    pd.testing.assert_index_equal(varp_result.index, minimal_adata.var.index)
+    pd.testing.assert_index_equal(varp_result.columns, pd.Index(varp_attrs))
+    np.testing.assert_array_equal(varp_result.values, varp_array)
+
+    # Test X table
+    obs_attrs = minimal_adata.obs.index[:3].tolist()  # First three observations
+    x_array = minimal_adata.X[0:3, :]  # Select first three observations
+    x_result = loading._create_results_df(
+        array=x_array,
+        attrs=obs_attrs,
+        var_index=minimal_adata.var.index,
+        table_type=ADATA.X,
+    )
+    assert x_result.shape == (minimal_adata.n_vars, len(obs_attrs))
+    pd.testing.assert_index_equal(x_result.index, minimal_adata.var.index)
+    pd.testing.assert_index_equal(x_result.columns, pd.Index(obs_attrs))
+    np.testing.assert_array_equal(x_result.values, x_array.T)
+
+    # Test layers table
+    layer_attrs = minimal_adata.obs.index[:2].tolist()  # First two observations
+    layer_array = minimal_adata.layers["counts"][
+        0:2, :
+    ]  # Select first two observations
+    layer_result = loading._create_results_df(
+        array=layer_array,
+        attrs=layer_attrs,
+        var_index=minimal_adata.var.index,
+        table_type=ADATA.LAYERS,
+    )
+    assert layer_result.shape == (minimal_adata.n_vars, len(layer_attrs))
+    pd.testing.assert_index_equal(layer_result.index, minimal_adata.var.index)
+    pd.testing.assert_index_equal(layer_result.columns, pd.Index(layer_attrs))
+    np.testing.assert_array_equal(layer_result.values, layer_array.T)
+
+
+@pytest.fixture
+def minimal_mudata(minimal_adata):
+    """Create a minimal MuData object for testing.
+
+    Uses minimal_adata as the RNA modality and creates a simple protein modality.
+    Focuses on testing MuData-level operations rather than modality-specific features.
+    """
+    # Create protein modality with minimal features
+    n_vars_protein = 3
+    adata_protein = anndata.AnnData(
+        X=np.random.randn(minimal_adata.n_obs, n_vars_protein),
+        obs=minimal_adata.obs,  # Share obs to ensure alignment
+        var=pd.DataFrame(
+            {
+                "uniprot": [
+                    f"P{i:05d}" for i in range(n_vars_protein)
+                ]  # Valid ontology column
+            },
+            index=[f"protein_{i}" for i in range(n_vars_protein)],
+        ),
+    )
+
+    # Create MuData with both modalities
+    mdata = mudata.MuData({"rna": minimal_adata, "protein": adata_protein})
+
+    # Add varm table at MuData level
+    n_features = 3
+    varm_array = np.random.randn(mdata.n_vars, n_features)
+    mdata.varm["gene_scores"] = varm_array
+    mdata.uns["gene_scores_features"] = ["score1", "score2", "score3"]
+
+    return mdata
+
+
+def test_prepare_anndata_results_df_anndata(minimal_adata):
+    """Test prepare_anndata_results_df with AnnData input."""
+    # Test var table
+    var_results = loading.prepare_anndata_results_df(
+        minimal_adata, table_type=ADATA.VAR
+    )
+    assert isinstance(var_results, pd.DataFrame)
+    assert var_results.shape[0] == minimal_adata.n_vars
+    assert "gene_name" in var_results.columns
+    assert "ensembl_transcript" in var_results.columns
+
+    # Test varm table
+    varm_results = loading.prepare_anndata_results_df(
+        minimal_adata,
+        table_type=ADATA.VARM,
+        table_name="gene_scores",
+        results_attrs=minimal_adata.uns["gene_scores_features"],
+        table_colnames=minimal_adata.uns["gene_scores_features"],  # Pass column names
+    )
+    assert isinstance(varm_results, pd.DataFrame)
+    assert varm_results.shape[0] == minimal_adata.n_vars
+    assert (
+        varm_results.shape[1] == 5
+    )  # score1, score2, score3 + gene_name + ensembl_transcript from var table
+
+    # Check we have both the scores and systematic identifiers
+    assert all(
+        score in varm_results.columns
+        for score in minimal_adata.uns["gene_scores_features"]
+    )
+    assert "gene_name" in varm_results.columns
+    assert "ensembl_transcript" in varm_results.columns
+
+    # Test with ontology extraction
+    var_results_with_ontology = loading.prepare_anndata_results_df(
+        minimal_adata,
+        table_type=ADATA.VAR,
+        index_which_ontology="ensembl_gene",  # Use a new ontology name
+    )
+    assert "ensembl_gene" in var_results_with_ontology.columns
+    pd.testing.assert_series_equal(
+        var_results_with_ontology["ensembl_gene"],
+        pd.Series(
+            minimal_adata.var.index, index=minimal_adata.var.index, name="ensembl_gene"
+        ),
+    )
+
+    # Test error when trying to use existing column
+    with pytest.raises(
+        ValueError, match="Cannot use 'gene_name' as index_which_ontology"
+    ):
+        loading.prepare_anndata_results_df(
+            minimal_adata,
+            table_type=ADATA.VAR,
+            index_which_ontology="gene_name",  # Should fail - already exists
+        )
+
+
+def test_split_mdata_results_by_modality(minimal_mudata):
+    """Test splitting results table by modality."""
+    # Create a combined results table with all vars
+    all_results = pd.DataFrame(
+        np.random.randn(len(minimal_mudata.var_names), 2),
+        index=minimal_mudata.var_names,
+        columns=["score1", "score2"],
+    )
+
+    # Split by modality
+    modality_results = loading._split_mdata_results_by_modality(
+        minimal_mudata, all_results
+    )
+
+    # Check we got both modalities
+    assert set(modality_results.keys()) == {"rna", "protein"}
+
+    # Check RNA results
+    rna_results = modality_results["rna"]
+    assert isinstance(rna_results, pd.DataFrame)
+    assert rna_results.shape[0] == minimal_mudata.mod["rna"].n_vars
+    assert list(rna_results.index) == list(minimal_mudata.mod["rna"].var.index)
+    assert list(rna_results.columns) == ["score1", "score2"]
+
+    # Check protein results
+    protein_results = modality_results["protein"]
+    assert isinstance(protein_results, pd.DataFrame)
+    assert protein_results.shape[0] == minimal_mudata.mod["protein"].n_vars
+    assert list(protein_results.index) == list(minimal_mudata.mod["protein"].var.index)
+    assert list(protein_results.columns) == ["score1", "score2"]
+
+    # Check that all original results are preserved
+    assert len(all_results) == sum(len(df) for df in modality_results.values())
+    for modality, results in modality_results.items():
+        pd.testing.assert_frame_equal(
+            results, all_results.loc[minimal_mudata.mod[modality].var_names]
+        )
+
+
+def test_split_mdata_results_by_modality_errors(minimal_mudata):
+    """Test error cases for splitting results by modality."""
+    # Create results with wrong index but matching length
+    wrong_index_results = pd.DataFrame(
+        np.random.randn(len(minimal_mudata.var_names), 2),
+        # Create completely different indices that don't overlap with real ones
+        index=[f"wrong_var_{i}" for i in range(len(minimal_mudata.var_names))],
+        columns=["score1", "score2"],
+    )
+
+    # Should raise error due to index mismatch
+    with pytest.raises(ValueError, match="Index mismatch in rna"):
+        loading._split_mdata_results_by_modality(minimal_mudata, wrong_index_results)
+
+
+def test_multimodality_ontology_config():
+    """Test MultiModalityOntologyConfig creation and validation."""
+    # Test successful creation with different ontology types
+    config_dict = {
+        "transcriptomics": {
+            "ontologies": None,  # Auto-detect
+            "index_which_ontology": None,
+        },
+        "proteomics": {
+            "ontologies": {"uniprot", "pharos"},  # Set of columns
+            "index_which_ontology": "uniprot",
+        },
+        "atac": {
+            "ontologies": {"peak1": "peak_id"},  # Dict mapping
+            "index_which_ontology": None,
+        },
+    }
+    config = loading.MultiModalityOntologyConfig.from_dict(config_dict)
+
+    # Test dictionary-like access
+    assert len(config) == 3
+    assert set(config) == {"transcriptomics", "proteomics", "atac"}
+
+    # Test modality access and type preservation
+    transcriptomics = config["transcriptomics"]
+    assert transcriptomics.ontologies is None
+    assert transcriptomics.index_which_ontology is None
+
+    proteomics = config["proteomics"]
+    assert isinstance(proteomics.ontologies, set)
+    assert proteomics.ontologies == {"uniprot", "pharos"}
+    assert proteomics.index_which_ontology == "uniprot"
+
+    atac = config["atac"]
+    assert isinstance(atac.ontologies, dict)
+    assert atac.ontologies == {"peak1": "peak_id"}
+    assert atac.index_which_ontology is None
+
+    # Test items() method
+    for modality, modality_config in config.items():
+        assert modality in config_dict
+        assert modality_config.ontologies == config_dict[modality]["ontologies"]
+        assert (
+            modality_config.index_which_ontology
+            == config_dict[modality]["index_which_ontology"]
+        )
+
+    # Test validation - missing required field
+    with pytest.raises(ValueError):
+        loading.MultiModalityOntologyConfig.from_dict(
+            {
+                "transcriptomics": {
+                    "index_which_ontology": "ensembl_gene"  # Missing ontologies field
+                }
+            }
+        )
+
+    # Test validation - wrong type for ontologies
+    with pytest.raises(ValueError):
+        loading.MultiModalityOntologyConfig.from_dict(
+            {
+                "transcriptomics": {
+                    "ontologies": "ensembl_gene",  # Should be None, set, or dict
+                    "index_which_ontology": "ensembl_gene",
+                }
+            }
+        )
+
+    # Test empty config
+    empty_config = loading.MultiModalityOntologyConfig(root={})
+    assert len(empty_config) == 0
+
+    # Test optional index_which_ontology
+    minimal_config = loading.MultiModalityOntologyConfig.from_dict(
+        {"transcriptomics": {"ontologies": None}}  # No index_which_ontology
+    )
+    assert minimal_config["transcriptomics"].index_which_ontology is None
+
+
+def test_prepare_mudata_results_df(minimal_mudata):
+    """Test prepare_mudata_results_df with different ontology configurations.
+
+    The function should:
+    1. Use MuData's var/varm tables for the actual data
+    2. Use each modality's var table for ontology information
+    3. Return a dictionary of DataFrames, one per modality
+    4. Each DataFrame should have the ontology columns and any requested data columns
+    """
+    # Arrange
+    config = {
+        "rna": {
+            "ontologies": None,  # Auto-detect
+            "index_which_ontology": "ensembl_gene",  # Rename index to this
+        },
+        "protein": {
+            "ontologies": {
+                "protein_id": "ensembl_protein",
+                "uniprot": "uniprot",
+            },  # Map source column to valid ontology name
+            "index_which_ontology": None,
+        },
+    }
+
+    # First add the source column to the protein modality
+    minimal_mudata.mod["protein"].var["protein_id"] = minimal_mudata.mod["protein"].var[
+        "uniprot"
+    ]
+
+    expected_rna_ontologies = {
+        "ensembl_gene",  # From index
+        "ensembl_transcript",  # From RNA modality var
+        "gene_name",  # From RNA modality var
+    }
+    expected_protein_ontologies = {
+        "uniprot",
+        "ensembl_protein",
+    }  # Both original and renamed columns
+
+    # Act - Test var table extraction
+    var_results = loading.prepare_mudata_results_df(
+        minimal_mudata, mudata_ontologies=config, table_type=ADATA.VAR
+    )
+
+    # Assert - Basic structure
+    assert set(var_results.keys()) == {"rna", "protein"}
+
+    # Assert - RNA modality
+    rna_results = var_results["rna"]
+    assert isinstance(rna_results, pd.DataFrame)
+    assert rna_results.shape[0] == minimal_mudata.mod["rna"].n_vars
+    assert expected_rna_ontologies.issubset(
+        set(rna_results.columns)
+    ), f"Missing ontology columns: {expected_rna_ontologies - set(rna_results.columns)}"
+
+    # Assert - Protein modality
+    protein_results = var_results["protein"]
+    assert isinstance(protein_results, pd.DataFrame)
+    assert protein_results.shape[0] == minimal_mudata.mod["protein"].n_vars
+    assert expected_protein_ontologies.issubset(
+        set(protein_results.columns)
+    ), f"Missing ontology columns: {expected_protein_ontologies - set(protein_results.columns)}"
+    # Check that source column was correctly renamed
+    assert "protein_id" not in protein_results.columns
+    assert "ensembl_protein" in protein_results.columns
+    pd.testing.assert_series_equal(
+        protein_results["ensembl_protein"],
+        minimal_mudata.mod["protein"].var["protein_id"],
+        check_names=False,  # Ignore Series names in comparison
+    )
+
+    # Act - Test varm table extraction with explicit results_attrs
+    varm_results = loading.prepare_mudata_results_df(
+        minimal_mudata,
+        mudata_ontologies=config,
+        table_type=ADATA.VARM,
+        table_name="gene_scores",
+        results_attrs=["score1", "score2"],
+        table_colnames=["score1", "score2", "score3"],
+    )
+
+    # Assert - RNA varm results
+    rna_varm = varm_results["rna"]
+    expected_rna_varm_cols = expected_rna_ontologies | {"score1", "score2"}
+    assert isinstance(rna_varm, pd.DataFrame)
+    assert rna_varm.shape[0] == minimal_mudata.mod["rna"].n_vars
+    assert expected_rna_varm_cols.issubset(
+        set(rna_varm.columns)
+    ), f"Missing columns: {expected_rna_varm_cols - set(rna_varm.columns)}"
+
+    # Assert - Protein varm results
+    protein_varm = varm_results["protein"]
+    expected_protein_varm_cols = expected_protein_ontologies | {"score1", "score2"}
+    assert isinstance(protein_varm, pd.DataFrame)
+    assert protein_varm.shape[0] == minimal_mudata.mod["protein"].n_vars
+    assert expected_protein_varm_cols.issubset(
+        set(protein_varm.columns)
+    ), f"Missing columns: {expected_protein_varm_cols - set(protein_varm.columns)}"
+
+
+def test_prepare_mudata_results_df_errors(minimal_mudata):
+    """Test error cases for prepare_mudata_results_df."""
+    # Test missing modality configuration
+    with pytest.raises(
+        ValueError, match="Missing ontology configurations for modalities"
+    ):
+        loading.prepare_mudata_results_df(
+            minimal_mudata,
+            mudata_ontologies={"rna": {"ontologies": None}},
+            table_type=ADATA.VAR,
+        )
+
+    # Test invalid table type
+    with pytest.raises(ValueError, match="table_type must be one of"):
+        loading.prepare_mudata_results_df(
+            minimal_mudata,
+            mudata_ontologies={
+                "rna": {"ontologies": None},
+                "protein": {"ontologies": None},
+            },
+            table_type="invalid_type",
+        )
+
+    # Test missing table_colnames for varm
+    # Add varm table to RNA modality first
+    minimal_mudata.mod["rna"].varm["scores"] = np.random.randn(
+        minimal_mudata.mod["rna"].n_vars, 3
+    )
+    minimal_mudata.mod["rna"].uns["scores_features"] = ["score1", "score2", "score3"]
+    with pytest.raises(ValueError, match="table_name 'scores' not found in adata.varm"):
+        loading.prepare_mudata_results_df(
+            minimal_mudata,
+            mudata_ontologies={
+                "rna": {"ontologies": None},
+                "protein": {"ontologies": None},
+            },
+            table_type=ADATA.VARM,
+            table_name="scores",
+            results_attrs=["score1", "score2"],  # Missing table_colnames
+        )
+
+
+def test_prepare_mudata_results_df_adata_level(minimal_mudata):
+    """Test prepare_mudata_results_df with level='adata' to extract adata-specific attributes."""
+    # Arrange - Add some adata-specific var attributes that don't exist at MuData level
+    minimal_mudata.mod["rna"].var["rna_specific"] = [
+        f"rna_val_{i}" for i in range(minimal_mudata.mod["rna"].n_vars)
+    ]
+    minimal_mudata.mod["protein"].var["protein_specific"] = [
+        f"prot_val_{i}" for i in range(minimal_mudata.mod["protein"].n_vars)
+    ]
+
+    config = {
+        "rna": {
+            "ontologies": None,  # Auto-detect
+            "index_which_ontology": None,
+        },
+        "protein": {
+            "ontologies": {"uniprot"},
+            "index_which_ontology": None,
+        },
+    }
+
+    # Act - Test var table extraction at adata level
+    var_results = loading.prepare_mudata_results_df(
+        minimal_mudata,
+        mudata_ontologies=config,
+        table_type=ADATA.VAR,
+        level=SCVERSE_DEFS.ADATA,
+    )
+
+    # Assert - Check that adata-specific attributes are included
+    rna_results = var_results["rna"]
+    protein_results = var_results["protein"]
+
+    # RNA should have its specific attribute
+    assert "rna_specific" in rna_results.columns
+    assert "gene_name" in rna_results.columns  # From original RNA var
+    assert "ensembl_transcript" in rna_results.columns  # From original RNA var
+
+    # Protein should have its specific attribute
+    assert "protein_specific" in protein_results.columns
+    assert "uniprot" in protein_results.columns  # From original protein var
+
+    # Check values are correct
+    pd.testing.assert_series_equal(
+        rna_results["rna_specific"],
+        minimal_mudata.mod["rna"].var["rna_specific"],
+        check_names=False,
+    )
+    pd.testing.assert_series_equal(
+        protein_results["protein_specific"],
+        minimal_mudata.mod["protein"].var["protein_specific"],
+        check_names=False,
+    )
+
+
+def test_prepare_mudata_results_df_mdata_vs_adata_level(minimal_mudata):
+    """Test that level='mdata' vs level='adata' produce different results when appropriate."""
+    # Arrange - Add adata-specific varm tables
+    rna_varm = np.random.randn(minimal_mudata.mod["rna"].n_vars, 2)
+    protein_varm = np.random.randn(minimal_mudata.mod["protein"].n_vars, 2)
+
+    minimal_mudata.mod["rna"].varm["modality_scores"] = rna_varm
+    minimal_mudata.mod["protein"].varm["modality_scores"] = protein_varm
+
+    config = {
+        "rna": {"ontologies": None},
+        "protein": {"ontologies": {"uniprot"}},
+    }
+
+    # Act - Extract using both levels
+    mdata_results = loading.prepare_mudata_results_df(
+        minimal_mudata,
+        mudata_ontologies=config,
+        table_type=ADATA.VARM,
+        table_name="gene_scores",  # This exists at MuData level
+        results_attrs=["score1"],
+        table_colnames=["score1", "score2", "score3"],
+        level=SCVERSE_DEFS.MDATA,
+    )
+
+    adata_results = loading.prepare_mudata_results_df(
+        minimal_mudata,
+        mudata_ontologies=config,
+        table_type=ADATA.VARM,
+        table_name="modality_scores",  # This exists at modality level
+        results_attrs=[
+            "0"
+        ],  # Using column index as string since we don't have explicit names
+        table_colnames=["0", "1"],
+        level=SCVERSE_DEFS.ADATA,
+    )
+
+    # Assert - Both should succeed but access different data
+    assert "score1" in mdata_results["rna"].columns
+    assert "score1" in mdata_results["protein"].columns
+
+    assert "0" in adata_results["rna"].columns
+    assert "0" in adata_results["protein"].columns
+
+    # The values should be different since they come from different varm tables
+    # (We can't easily check exact values due to random generation, but structure should be correct)
+    assert mdata_results["rna"].shape[0] == minimal_mudata.mod["rna"].n_vars
+    assert adata_results["rna"].shape[0] == minimal_mudata.mod["rna"].n_vars
+
+
+def test_prepare_mudata_results_df_level_validation(minimal_mudata):
+    """Test that invalid level parameter raises appropriate error."""
+    config = {
+        "rna": {"ontologies": None},
+        "protein": {"ontologies": {"uniprot"}},
+    }
+
+    with pytest.raises(
+        ValueError,
+        match=r"level must be one of \['adata', 'mdata'\], got invalid_level",
+    ):
+        loading.prepare_mudata_results_df(
+            minimal_mudata,
+            mudata_ontologies=config,
+            table_type=ADATA.VAR,
+            level="invalid_level",
+        )