rectanglepy 0.1.6__tar.gz → 0.1.8__tar.gz

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.1.6
+current_version = 0.1.8
 tag = True
 commit = True
 

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/CHANGELOG.md

@@ -10,6 +10,22 @@ and this project adheres to [Semantic Versioning][].
 
 ## [Unreleased]
 
+## [0.1.8] - 2024-09-06
+
+### Added
+
+-   RectangleSignatureResult now returns correlation of bulk genes with the calculated Unknown cell type fraction
+
+### Fixed
+
+-   The consensus method now returns the estimations normalized to 1
+
+## [0.1.7] - 2024-07-25
+
+### Fixed
+
+-   Fixed bug in the consensus method.
+
 ## [0.1.6] - 2024-05-07
 
 ### Added

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: rectanglepy
-Version: 0.1.6
+Version: 0.1.8
 Summary: Hierarchical deconvolution of bulk  transcriptomics
 Project-URL: Documentation, https://rectanglepy.readthedocs.io/
 Project-URL: Source, https://github.com/ComputationalBiomedicineGroup/Rectangle
@@ -30,7 +30,9 @@ License: MIT License
         SOFTWARE.
 License-File: LICENSE
 Requires-Python: >=3.10
+Requires-Dist: anndata<0.10.9,>=0.8.0
 Requires-Dist: loguru
+Requires-Dist: numpy<2.0.0,>=1.0.0
 Requires-Dist: pydeseq2==0.4.1
 Requires-Dist: quadprog==0.1.11
 Provides-Extra: dev

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/pyproject.toml

@@ -4,7 +4,7 @@ requires = ["hatchling"]
 
 [project]
 name = "rectanglepy"
-version = "0.1.6"
+version = "0.1.8"
 description = "Hierarchical deconvolution of bulk  transcriptomics"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -22,6 +22,8 @@ dependencies = [
     "pydeseq2==0.4.1",
     "quadprog==0.1.11",
     "loguru",
+    "numpy>=1.0.0,<2.0.0",
+    "anndata>=0.8.0,<0.10.9",
 ]
 
 [project.optional-dependencies]

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/src/rectanglepy/pp/create_signature.py

@@ -34,7 +34,9 @@ def _create_condition_number_matrices(de_adjusted, pseudo_signature):
     loop_range = min(longest_de_analysis, 200)
     range_minimum = 30
 
-    if loop_range < range_minimum:
+    if loop_range < 8:
+        range_minimum = 2
+    elif loop_range < range_minimum:
         range_minimum = 8
 
     for i in range(range_minimum, loop_range):
@@ -245,6 +247,7 @@ def build_rectangle_signatures(
     subsample: bool = False,
     sample_size: int = 1500,
     n_cpus: int = None,
+    run: int = 0,
 ) -> RectangleSignatureResult:
     r"""Builds rectangle signatures based on single-cell  count data and annotations.
 
@@ -272,6 +275,8 @@ def build_rectangle_signatures(
         The log fold change threshold for the DE analysis (only used if optimize_cutoffs is False).
     n_cpus
         The number of cpus to use for the DE analysis. Defaults to the number of cpus available.
+    run
+        The consensus run number for the analysis. Defaults to 0.
 
     Returns
     -------
@@ -286,7 +291,7 @@ def build_rectangle_signatures(
 
     annotations = adata.obs[cell_type_col]
     if subsample:
-        annotations = _even(annotations, sample_size)
+        annotations = _even(annotations, sample_size, run)
         adata = adata[annotations.index]
 
     if layer is not None:
@@ -365,9 +370,14 @@ def _optimize_parameters(
     bulks, real_fractions = _generate_pseudo_bulks(sc_data, annotations, genes)
     for p in ps:
         for lfc in lfcs:
-            rmse, pearson_r = _assess_parameter_fit(lfc, p, bulks, real_fractions, pseudo_signature_counts, de_results)
-            logger.info(f"RMSE:{rmse}, Pearson R:{pearson_r} for p={p}, lfc={lfc}")
-            results.append({"p": p, "lfc": lfc, "rmse": rmse, "pearson_r": pearson_r})
+            try:
+                rmse, pearson_r = _assess_parameter_fit(
+                    lfc, p, bulks, real_fractions, pseudo_signature_counts, de_results
+                )
+                logger.info(f"RMSE:{rmse}, Pearson R:{pearson_r} for p={p}, lfc={lfc}")
+                results.append({"p": p, "lfc": lfc, "rmse": rmse, "pearson_r": pearson_r})
+            except Exception as e:
+                logger.error(f"Error in assessing parameter fit for p={p}, lfc={lfc}: {e}")
 
     results_df = pd.DataFrame(results)
     results_df = results_df.sort_values(by=["pearson_r", "rmse"], ascending=[False, True])
@@ -448,7 +458,8 @@ def _reduce_to_common_genes(bulks: pd.DataFrame, sc_data: pd.DataFrame):
     return bulks, sc_data
 
 
-def _even(annotations: pd.Series, number: int) -> pd.Series:
+def _even(annotations: pd.Series, number: int, run=0) -> pd.Series:
+    np.random.seed(run)
     assert number > 0, "Number of cells must be greater than 0"
     annotation_counts = annotations.value_counts()
     selected_cells = []

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/src/rectanglepy/pp/rectangle_signature.py

@@ -22,6 +22,8 @@ class RectangleSignatureResult:
         The number of marker genes per cell type, as a dictionary. Keys are cell type names, values are the number of marker genes.
     optimization_result
         The result of the p lfc cut off optimization, as a pd.DataFrame. Contains the following columns: p, lfc, pearson_r, rsme
+    unkn_gene_corr
+        The correlation between the unknown cell type and the genes linked to the unknown cell type.
     """
 
     def __init__(
@@ -36,6 +38,7 @@ class RectangleSignatureResult:
         marker_genes_per_cluster: dict[str, [str]] = None,
         clustered_signature_genes: pd.Series = None,
         cluster_assignments: list[int or str] = None,
+        unkn_gene_corr: pd.Series = None,
     ):
         self.signature_genes = signature_genes
         self.bias_factors = bias_factors
@@ -47,6 +50,7 @@ class RectangleSignatureResult:
         self.clustered_bias_factors = clustered_bias_factors
         self.clustered_signature_genes = clustered_signature_genes
         self.assignments = cluster_assignments
+        self.unkn_gene_corr = unkn_gene_corr
 
     def get_signature_matrix(self, include_mrna_bias=True) -> pd.DataFrame:
         """Calculates the signature matrix by multiplying the pseudobulk_sig_cpm DataFrame subset by signature_genes and the bias_factors Series.

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/src/rectanglepy/rectangle.py

@@ -94,8 +94,8 @@ def rectangle_consens(
     rectangle_signature_results = []
     most_recent_signatures = None
 
-    for _i in range(consensus_runs):
-        logger.info(f"Running consensus run {_i + 1} of {consensus_runs}")
+    for i in range(consensus_runs):
+        logger.info(f"Running consensus run {i + 1} of {consensus_runs}")
         signatures = build_rectangle_signatures(
             adata,
             cell_type_col,
@@ -108,14 +108,19 @@ def rectangle_consens(
             n_cpus=n_cpus,
             subsample=subsample,
             sample_size=sample_size,
+            run=i,
         )
-        rectangle_signature_results.append(signatures)
         most_recent_signatures = signatures
         cell_fractions = deconvolution(signatures, bulks, correct_mrna_bias, n_cpus)
         estimations.append(cell_fractions)
-
+        unkn_gene_corr = _genes_linked_to_unkn(bulks, cell_fractions["Unknown"])
+        signatures.unkn_gene_corr = unkn_gene_corr
+        rectangle_signature_results.append(signatures)
+    consensus_estimations = pd.concat(estimations).groupby(level=0).median()
+    # normalize the estimations to 1
+    consensus_estimations = consensus_estimations.div(consensus_estimations.sum(axis=1), axis=0)
     consensus_results = ConsensusResult(estimations, rectangle_signature_results)
-    return pd.concat(estimations).groupby(level=0).median(), most_recent_signatures, consensus_results
+    return consensus_estimations, most_recent_signatures, consensus_results
 
 
 def rectangle(
@@ -196,3 +201,11 @@ def load_tutorial_data() -> tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame]:
         bulks = pd.read_csv(bulks_file, index_col=0, compression="zip")
 
     return sc_counts.T, annotations, bulks.T
+
+
+def _genes_linked_to_unkn(bulks: DataFrame, unkn_fractions: pd.Series):
+    genes = bulks.columns
+    corr = []
+    for gene in genes:
+        corr.append(unkn_fractions.corr(bulks.loc[:, gene]))
+    return pd.Series(corr, index=genes).sort_values(ascending=False)

{rectanglepy-0.1.6 → rectanglepy-0.1.8}/tests/test_rectangle.py

@@ -15,22 +15,37 @@ def test_load_tutorial_data():
 
 def test_rectangle():
     sc_data, annotations, bulks = rectanglepy.load_tutorial_data()
+    sc_data = sc_data.iloc[:, :500]
     sc_data_adata = AnnData(sc_data, obs=annotations.to_frame(name="cell_type"))
 
     result = rectanglepy.rectangle(sc_data_adata, bulks)
+    estimations, signatures = result
 
-    assert isinstance(result[0], pd.DataFrame)
-    assert isinstance(result[1], RectangleSignatureResult)
+    assert isinstance(estimations, pd.DataFrame)
+    assert isinstance(signatures, RectangleSignatureResult)
+    assert isinstance(signatures.unkn_gene_corr, pd.Series)
 
 
 def test_rectangle_consensus():
     sc_data, annotations, bulks = rectanglepy.load_tutorial_data()
+    sc_data = sc_data.iloc[:, :500]
     sc_data_adata = AnnData(sc_data, obs=annotations.to_frame(name="cell_type"))
 
     result = rectanglepy.rectangle_consens(
-        sc_data_adata, bulks, optimize_cutoffs=False, p=0.5, lfc=0.0, consensus_runs=2, sample_size=50
+        sc_data_adata, bulks, optimize_cutoffs=False, p=0.5, lfc=0.0, consensus_runs=3, sample_size=50
     )
 
     assert isinstance(result[0], pd.DataFrame)
     assert isinstance(result[1], RectangleSignatureResult)
     assert isinstance(result[2], ConsensusResult)
+
+    estimations = result[0]
+    consensus_results = result[2]
+    signature_results = consensus_results.rectangle_signature_results
+    bias_factors = [result.bias_factors for result in signature_results]
+    # there was a problem with the random sampling, the bias factors should differ
+    assert bias_factors[0]["Monocytes"] != bias_factors[1]["Monocytes"]
+    assert bias_factors[0]["Monocytes"] != bias_factors[2]["Monocytes"]
+    assert bias_factors[1]["Monocytes"] != bias_factors[2]["Monocytes"]
+    # in the consensus we have to make sure that the estimations are again normalized to 1
+    assert estimations.sum(axis=1).all() == 1.0