MAIN fix issue 387 (#394)

* refactor: move 1D data to var or obs

* refactor: use df.loc[row_indexer, "col"] syntax instead of df["col"][row_indexer]

* refactor: use df.loc[row_indexer, "col"] syntax instead of df["col"][row_indexer]

* fix: pass arguments as numpy arrays instead of pandas Series to avoid integer indexing issues

* refactor: use df.loc[row_indexer, "col"] syntax instead of df["col"][row_indexer]

* test: update anndata fields

* refactor: update indexing

* fix: pass dispersions as numpy array (instead of pd Series) in wald tests

* fix: cast size_factors to numpy array in _replace_outliers()

* fix: convert to numpy array before using [:, None]

* fix: convert dispersions to numpy array before calling nb_nll()

* tests: remove unnecessary ravel() calls

* fix: fix ChainedAssignmentError

* fix: tests and indexing

* tests: remove breakpoint

* docs: update fields in examples

Author: BorisMuzellec

examples/plot_minimal_pydeseq2_pipeline.py

@@ -192,18 +192,18 @@
 # structure, with key-based data fields. In particular,
 #
 # - ``X`` stores the count data,
-# - ``obs`` stores design factors,
-# - ``obsm`` stores sample-level data, such as ``"design_matrix"`` and
-#   ``"size_factors"``,
-# - ``varm`` stores gene-level data, such as ``"dispersions"`` and ``"LFC"``.
+# - ``obs`` stores 1D sample-level data, such as design factors and ``"size_factors"``,
+# - ``obsm`` stores multi-dimensional sample-level data, such as ``"design_matrix"``,
+# - ``var`` stores 1D gene-level data, such as gene names and ``"dispersions"``,
+# - ``varm`` stores multi-dimensional gene-level data, such as ``"LFC"``.
 #
 #
 # As an example, here is how we would access dispersions and LFCs
 # (in natural log scale):
 
 # %%
 
-print(dds.varm["dispersions"])
+print(dds.var["dispersions"])
 
 # %%
 

examples/plot_pandas_io_example.py

@@ -161,18 +161,17 @@
 # structure, with key-based data fields. In particular,
 #
 # - ``X`` stores the count data,
-# - ``obs`` stores design factors,
-# - ``obsm`` stores sample-level data, such as ``"design_matrix"`` and
-#   ``"size_factors"``,
-# - ``varm`` stores gene-level data, such as ``"dispersions"`` and ``"LFC"``.
-#
+# - ``obs`` stores 1D sample-level data, such as design factors and ``"size_factors"``,
+# - ``obsm`` stores multi-dimensional sample-level data, such as ``"design_matrix"``,
+# - ``var`` stores 1D gene-level data, such as gene names and ``"dispersions"``,
+# - ``varm`` stores multi-dimensional gene-level data, such as ``"LFC"``.
 #
 # As an example, here is how we would access dispersions and LFCs
 # (in natural log scale):
 
 # %%
 
-print(dds.varm["dispersions"])
+print(dds.var["dispersions"])
 
 # %%
 

examples/plot_step_by_step.py

@@ -107,23 +107,23 @@
 
 dds.fit_size_factors()
 
-dds.obsm["size_factors"]
+dds.obs["size_factors"]
 
 # %%
 # Fit genewise dispersions
 # ^^^^^^^^^^^^^^^^^^^^^^^^
 
 dds.fit_genewise_dispersions()
 
-dds.varm["genewise_dispersions"]
+dds.var["genewise_dispersions"]
 
 # %%
 # Fit dispersion trend coefficients
 # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 dds.fit_dispersion_trend()
 dds.uns["trend_coeffs"]
-dds.varm["fitted_dispersions"]
+dds.var["fitted_dispersions"]
 
 # %%
 # Dispersion priors
@@ -145,8 +145,8 @@
 # stored in `dds.dispersions`.
 
 dds.fit_MAP_dispersions()
-dds.varm["MAP_dispersions"]
-dds.varm["dispersions"]
+dds.var["MAP_dispersions"]
+dds.var["dispersions"]
 
 # %%
 # Fit log fold changes