Refactored return data for all methods

Author: PauBadiaM

decoupler/__init__.py

@@ -1,7 +1,7 @@
 __version__ = '1.6.3'  # noqa: F401
 __version_info__ = tuple([int(num) for num in __version__.split('.')])  # noqa: F401
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n, mask_features  # noqa: F401
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, mask_features, return_data  # noqa: F401
 from .utils import (
     melt, show_methods, check_corr, get_toy_data, summarize_acts, assign_groups, dense_run, p_adjust_fdr, shuffle_net,
     read_gmt  # noqa: F401

decoupler/method_aucell.py

@@ -10,7 +10,7 @@
 from numpy.random import default_rng
 from tqdm import tqdm
 
-from .pre import extract, rename_net, filt_min_n
+from .pre import extract, rename_net, filt_min_n, return_data
 
 from anndata import AnnData
 import numba as nb
@@ -151,9 +151,4 @@ def run_aucell(mat, net, source='source', target='target', n_up=None, min_n=5, s
     estimate = pd.DataFrame(estimate, index=r, columns=net.index)
     estimate.name = 'aucell_estimate'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-    else:
-        return estimate
+    return return_data(mat=mat, results=(estimate, ))

decoupler/method_gsea.py

@@ -9,7 +9,7 @@
 from numpy.random import default_rng
 from scipy.sparse import csr_matrix
 
-from .pre import extract, rename_net, filt_min_n
+from .pre import extract, rename_net, filt_min_n, return_data
 from .utils import p_adjust_fdr
 
 from anndata import AnnData
@@ -369,15 +369,4 @@ def run_gsea(mat, net, source='source', target='target', times=1000, batch_size=
         pvals = pd.DataFrame(pvals, index=r, columns=net.index)
         pvals.name = 'gsea_pvals'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        if norm_e is not None:
-            mat.obsm[norm_e.name] = norm_e
-            mat.obsm[pvals.name] = pvals
-    else:
-        if pvals is not None:
-            return estimate, norm_e, pvals
-        else:
-            return estimate
+    return return_data(mat=mat, results=(estimate, norm_e, pvals))

decoupler/method_gsva.py

@@ -10,7 +10,7 @@
 from scipy.sparse import csr_matrix
 from numpy.random import default_rng
 
-from .pre import extract, rename_net, filt_min_n
+from .pre import extract, rename_net, filt_min_n, return_data
 from .method_gsea import std
 
 from anndata import AnnData
@@ -232,9 +232,4 @@ def run_gsva(mat, net, source='source', target='target', kcdf=False, mx_diff=Tru
     estimate = pd.DataFrame(estimate, index=r, columns=net.index)
     estimate.name = 'gsva_estimate'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-    else:
-        return estimate
+    return return_data(mat=mat, results=(estimate, ))

decoupler/method_mdt.py

@@ -7,7 +7,7 @@
 import pandas as pd
 from scipy.sparse import csr_matrix
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
 from anndata import AnnData
 from tqdm import tqdm
@@ -117,9 +117,4 @@ def run_mdt(mat, net, source='source', target='target', weight='weight', trees=1
     estimate = pd.DataFrame(estimate, index=r, columns=sources)
     estimate.name = 'mdt_estimate'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-    else:
-        return estimate
+    return return_data(mat=mat, results=(estimate, ))

decoupler/method_mlm.py

@@ -7,7 +7,7 @@
 import pandas as pd
 from scipy.sparse import csr_matrix
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
 from anndata import AnnData
 from scipy import stats
@@ -131,10 +131,4 @@ def run_mlm(mat, net, source='source', target='target', weight='weight', batch_s
     pvals = pd.DataFrame(pvals, index=r, columns=sources)
     pvals.name = 'mlm_pvals'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        mat.obsm[pvals.name] = pvals
-    else:
-        return estimate, pvals
+    return return_data(mat=mat, results=(estimate, pvals))

decoupler/method_ora.py

@@ -12,7 +12,7 @@
 from scipy.stats import rankdata
 from math import log, exp, lgamma
 
-from .pre import extract, rename_net, filt_min_n
+from .pre import extract, rename_net, filt_min_n, return_data
 from .utils import p_adjust_fdr
 
 from anndata import AnnData
@@ -315,10 +315,4 @@ def run_ora(mat, net, source='source', target='target', n_up=None, n_bottom=0, n
     estimate = pd.DataFrame(-np.log10(pvals), index=r, columns=pvals.columns)
     estimate.name = 'ora_estimate'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        mat.obsm[pvals.name] = pvals
-    else:
-        return estimate, pvals
+    return return_data(mat=mat, results=(estimate, pvals))

decoupler/method_udt.py

@@ -7,7 +7,7 @@
 from scipy.sparse import csr_matrix
 import pandas as pd
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
 from anndata import AnnData
 from tqdm import tqdm
@@ -114,9 +114,4 @@ def run_udt(mat, net, source='source', target='target', weight='weight', min_lea
     estimate = pd.DataFrame(estimate, index=r, columns=sources)
     estimate.name = 'udt_estimate'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-    else:
-        return estimate
+    return return_data(mat=mat, results=(estimate, ))

decoupler/method_ulm.py

@@ -9,7 +9,7 @@
 
 from scipy.stats import t
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
 from anndata import AnnData
 from tqdm import tqdm
@@ -124,10 +124,4 @@ def run_ulm(mat, net, source='source', target='target', weight='weight', batch_s
     pvals = pd.DataFrame(pvals, index=r, columns=sources)
     pvals.name = 'ulm_pvals'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        mat.obsm[pvals.name] = pvals
-    else:
-        return estimate, pvals
+    return return_data(mat=mat, results=(estimate, pvals))

decoupler/method_viper.py

@@ -10,7 +10,7 @@
 from scipy.stats import rankdata
 from scipy.stats import norm
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
 from anndata import AnnData
 from tqdm import tqdm
@@ -308,10 +308,4 @@ def run_viper(mat, net, source='source', target='target', weight='weight', pleio
     pvals = pd.DataFrame(pvals, index=r, columns=sources)
     pvals.name = 'viper_pvals'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        mat.obsm[pvals.name] = pvals
-    else:
-        return estimate, pvals
+    return return_data(mat=mat, results=(estimate, pvals))

decoupler/method_wmean.py

@@ -7,7 +7,7 @@
 import pandas as pd
 from scipy.sparse import csr_matrix
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 from .method_gsea import std
 
 from anndata import AnnData
@@ -177,16 +177,4 @@ def run_wmean(mat, net, source='source', target='target', weight='weight', times
         pvals = pd.DataFrame(pvals, index=r, columns=sources)
         pvals.name = 'wmean_pvals'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        if pvals is not None:
-            mat.obsm[norm.name] = norm
-            mat.obsm[corr.name] = corr
-            mat.obsm[pvals.name] = pvals
-    else:
-        if pvals is not None:
-            return estimate, norm, corr, pvals
-        else:
-            return estimate
+    return return_data(mat=mat, results=(estimate, norm, corr, pvals))

decoupler/method_wsum.py

@@ -7,7 +7,7 @@
 import pandas as pd
 from scipy.sparse import csr_matrix
 
-from .pre import extract, match, rename_net, get_net_mat, filt_min_n
+from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 from .method_gsea import std
 
 from anndata import AnnData
@@ -173,16 +173,4 @@ def run_wsum(mat, net, source='source', target='target', weight='weight', times=
         pvals = pd.DataFrame(pvals, index=r, columns=sources)
         pvals.name = 'wsum_pvals'
 
-    # AnnData support
-    if isinstance(mat, AnnData):
-        # Update obsm AnnData object
-        mat.obsm[estimate.name] = estimate
-        if pvals is not None:
-            mat.obsm[norm.name] = norm
-            mat.obsm[corr.name] = corr
-            mat.obsm[pvals.name] = pvals
-    else:
-        if pvals is not None:
-            return estimate, norm, corr, pvals
-        else:
-            return estimate
+    return return_data(mat=mat, results=(estimate, norm, corr, pvals))

decoupler/pre.py

@@ -6,7 +6,7 @@
 import numpy as np
 from scipy.sparse import csr_matrix, issparse
 import pandas as pd
-
+import logging
 from anndata import AnnData
 
 
@@ -278,3 +278,23 @@ def mask_features(mat, log=False, thr=1, use_raw=False):
     else:
         raise ValueError("""mat must be a list of [matrix, samples, features], dataframe (samples x features) or an AnnData
         instance.""")
+
+
+def add_to_anndata(mat, results):
+    for result in results:
+        if result is not None:
+            mat.obsm[result.name] = result
+
+
+def return_data(mat, results):
+    if isinstance(mat, AnnData):
+        if mat.obs_names.size != results[0].index.size:
+            logging.warning('Provided AnnData contains empty observations. Returning repaired object.')
+            mat = mat[results[0].index, :].copy()
+            add_to_anndata(mat, results)
+            return mat
+        else:
+            add_to_anndata(mat, results)
+            return None
+    else:
+        return tuple([result for result in results if result is not None])

decoupler/tests/test_pre.py

@@ -3,7 +3,10 @@
 import numpy as np
 from scipy.sparse import csr_matrix
 from anndata import AnnData
-from ..pre import check_mat, extract, filt_min_n, match, rename_net, get_net_mat, mask_features
+from ..pre import (
+    check_mat, extract, filt_min_n, match, rename_net, get_net_mat, mask_features,
+    return_data, add_to_anndata
+)
 
 
 def test_check_mat():
@@ -101,3 +104,38 @@ def test_mask_features():
         mask_features('asdfg')
     with pytest.raises(ValueError):
         mask_features(adata, use_raw=True)
+
+
+def test_add_to_anndata():
+    m = np.array([[1, 0, 2], [1, 0, 3]])
+    r = np.array(['S1', 'S2'])
+    c = np.array(['G1', 'G2', 'G3'])
+    df = pd.DataFrame(m, index=r, columns=c)
+    adata = AnnData(df.astype(np.float32))
+    estimate = np.array([[1], [4]])
+    s = np.array(['S1'])
+    estimate = pd.DataFrame(estimate, index=r, columns=s)
+    estimate.name = 'estimate'
+    add_to_anndata(mat=adata, results=(estimate, None))
+    assert 'estimate' in adata.obsm
+
+
+def test_return_data():
+    m = np.array([[1, 0, 2], [1, 0, 3], [0, 0, 0]])
+    r = np.array(['S1', 'S2', 'S3'])
+    c = np.array(['G1', 'G2', 'G3'])
+    df = pd.DataFrame(m, index=r, columns=c)
+    adata = AnnData(df.astype(np.float32))
+    estimate = np.array([[1], [4]])
+    s = np.array(['S1'])
+    estimate = pd.DataFrame(estimate, index=r[:-1], columns=s)
+    estimate.name = 'estimate'
+    pvals = np.array([[0.4], [0.01]])
+    pvals = pd.DataFrame(pvals, index=estimate.index, columns=estimate.columns)
+    pvals.name = 'pvals'
+    ret = return_data(mat=adata, results=(estimate, pvals))
+    assert isinstance(ret, AnnData)
+    ret = return_data(mat=adata[estimate.index, :].copy(), results=(estimate, pvals))
+    assert ret is None
+    ret = return_data(mat=df, results=(estimate, pvals))
+    assert isinstance(ret, tuple)

docs/source/release_notes.rst

@@ -9,10 +9,12 @@ Bug fixes
 - Fixed error in in ``get_contrast`` by reverting use of ``copy.deepcopy`` to ``copy``.
 - Fixed verbose error regarding the number of unique sources being used in ``benchmark``.
 - Added check for minimum version of ``igraph>=0.10.0`` to properly render ``plot_network``.
+- Fixed return error of methods triggered when an observation was empty and input was ``AnnData``.
 
 Changes
 ~~~~~~~
 - Resource functions such as ``get_resource`` or ``get_collectri`` now accept different ``genesymbol_resource`` than UniProt for gene translation to other organisms.
+- Deprecated ``sklearn`` and switched to ``sklearn`` for ``udt``.
 
 Additions
 ~~~~~~~~~