[ENH] Load data: Add data loader for H5AD format

orangecontrib/single_cell/tests/test_load_data.py

@@ -9,7 +9,7 @@
 
 from orangecontrib.single_cell.widgets.load_data import (
     LoomLoader, ExcelLoader, MtxLoader, CountLoader, Loader, PickleLoader,
-    CsvLoader, get_data_loader, Concatenate
+    CsvLoader, H5ADLoader, get_data_loader, Concatenate,
 )
 
 
@@ -21,6 +21,8 @@ def test_get_data_loader(self):
         self.assertIsInstance(loader, Loader)
         self.assertIsInstance(get_data_loader("data.xls"), ExcelLoader)
         self.assertIsInstance(get_data_loader("data.loom"), LoomLoader)
+        self.assertIsInstance(get_data_loader("data_sparse.h5ad"), H5ADLoader)
+        self.assertIsInstance(get_data_loader("data_dense.h5ad"), H5ADLoader)
 
     def test_get_data_loader_pickle(self):
         self.assertIsInstance(get_data_loader("data.pkl"), PickleLoader)
@@ -89,6 +91,22 @@ def test_file_summary_loom(self):
         self.assertEqual(loader.n_cols, 20)
         self.assertEqual(round(loader.sparsity, 2), 0.93)
 
+    def test_file_summary_h5ad_sparse(self):
+        file_name = os.path.join(os.path.dirname(__file__), "data/data_sparse.h5ad")
+        loader = H5ADLoader(file_name)
+        self.assertEqual(19560, loader.file_size)
+        self.assertEqual(20, loader.n_rows)
+        self.assertEqual(25, loader.n_cols)
+        self.assertEqual(0.59, round(loader.sparsity, 2))
+
+    def test_file_summary_h5ad_dense(self):
+        file_name = os.path.join(os.path.dirname(__file__), "data/data_dense.h5ad")
+        loader = H5ADLoader(file_name)
+        self.assertEqual(11060, loader.file_size)
+        self.assertEqual(20, loader.n_rows)
+        self.assertEqual(25, loader.n_cols)
+        self.assertEqual(0.59, round(loader.sparsity, 2))
+
     def test_load_data_mtx(self):
         file_name = os.path.join(os.path.dirname(__file__),
                                  "data/10x/mm10/matrix.mtx")

orangecontrib/single_cell/tests/test_owloaddata.py

@@ -5,7 +5,6 @@
 import pandas as pd
 
 from Orange.data import ContinuousVariable, Variable
-from Orange.widgets.data.owtable import OWDataTable
 from Orange.widgets.tests.base import WidgetTest
 
 from orangecontrib.single_cell.widgets.owloaddata import OWLoadData
@@ -241,6 +240,46 @@ def test_load_data_loom_sample(self):
         self.assertListEqual([attr.name for attr in data.domain.attributes],
                              ["0", "1", "2", "3", "6", "7"])
 
+    def test_load_data_h5ad_sparse_sample(self):
+        file_name = os.path.join(self._path, "data_sparse.h5ad")
+        self.widget.set_current_path(file_name)
+        self.widget.sample_rows_cb.setChecked(True)
+        self.widget.sample_cols_cb.setChecked(True)
+        self.widget.set_sample_rows_p(30)
+        self.widget.set_sample_cols_p(10)
+        self.widget.commit()
+        data = self.get_output("Data")
+        X = np.array([[0, 2, 1, 0], [1, 0, 0, 0], [2, 1, 2, 1], [1, 1, 0, 0],
+                      [1, 0, 0, 0], [0, 1, 0, 2], [1, 0, 0, 1], [0, 2, 0, 1],
+                      [0, 1, 1, 0]])
+        npt.assert_array_equal(data.X, X)
+        metas = np.array([[0, 0, 10], [1, 0, 10], [2, 0, 15], [3, 0, 10],
+                          [6, 1, 11], [7, 1, 15], [8, 1, 11], [9, 1, 20],
+                          [14, 2, 12]])
+        npt.assert_array_equal(data.metas, metas)
+        self.assertListEqual([attr.name for attr in data.domain.attributes],
+                             ["Gene 0", "Gene 1", "Gene 2", "Gene 3"])
+
+    def test_load_data_h5ad_dense_sample(self):
+        file_name = os.path.join(self._path, "data_dense.h5ad")
+        self.widget.set_current_path(file_name)
+        self.widget.sample_rows_cb.setChecked(True)
+        self.widget.sample_cols_cb.setChecked(True)
+        self.widget.set_sample_rows_p(30)
+        self.widget.set_sample_cols_p(10)
+        self.widget.commit()
+        data = self.get_output("Data")
+        X = np.array([[0, 2, 1, 0], [1, 0, 0, 0], [2, 1, 2, 1], [1, 1, 0, 0],
+                      [1, 0, 0, 0], [0, 1, 0, 2], [1, 0, 0, 1], [0, 2, 0, 1],
+                      [0, 1, 1, 0]])
+        npt.assert_array_equal(data.X, X)
+        metas = np.array([[0, 0, 10], [1, 0, 10], [2, 0, 15], [3, 0, 10],
+                          [6, 1, 11], [7, 1, 15], [8, 1, 11], [9, 1, 20],
+                          [14, 2, 12]])
+        npt.assert_array_equal(data.metas, metas)
+        self.assertListEqual([attr.name for attr in data.domain.attributes],
+                             ["Gene 0", "Gene 1", "Gene 2", "Gene 3"])
+
     def test_not_enough_headers(self):
         file_name = os.path.join(self._path, "DATA_MATRIX_LOG_TPM.txt")
         self.widget.set_current_path(file_name)

orangecontrib/single_cell/widgets/load_data.py

@@ -6,11 +6,12 @@
 import numpy as np
 import pandas as pd
 import scipy.io
+import scipy.sparse as sp
 import loompy as lp
 import xlrd
 
 from Orange.data import (
-    ContinuousVariable, DiscreteVariable, Domain, Table, StringVariable
+    ContinuousVariable, DiscreteVariable, Domain, Table
 )
 from Orange.data.io import (
     Compression, open_compressed, PickleReader,
@@ -49,6 +50,8 @@ def get_data_loader(file_name):
         return ExcelLoader(file_name)
     elif ext == ".loom":
         return LoomLoader(file_name)
+    elif ext == ".h5ad":
+        return H5ADLoader(file_name)
     else:
         return Loader(file_name)
 
@@ -425,7 +428,7 @@ def __into_orange_table(self, attrs, X, meta_parts):
     @staticmethod
     def __guess_metas(meta_parts):
         def guessed_var(i, var_name, dtype):
-            if np.issubdtype(dtype, np.number):
+            if pd.core.dtypes.common.is_numeric_dtype(dtype):
                 return ContinuousVariable.make(var_name)
             orig_values = M[:, i]
             val_map, values, var_type = guess_data_type(orig_values)
@@ -666,6 +669,57 @@ def _load_data(self, skip_row=None, skip_col=None, **kwargs):
             attrs = [ContinuousVariable.make(str(g)) for g in gene_names]
             meta_df = pd.DataFrame({key: ds.ca[key][self._use_cols_mask]
                                     for key in ds.ca.keys()})
+
+        return attrs, X, meta_df, meta_df.index
+
+
+class H5ADLoader(Loader):
+    def __init__(self, file_name):
+        super().__init__(file_name)
+        self.header_rows_count = 0
+        self.header_cols_count = 0
+        self.FIXED_FORMAT = False
+        self.ENABLE_ANNOTATIONS = False
+        self.transposed = False
+        self.row_annotations_enabled = False
+        self.col_annotations_enabled = False
+
+    def _set_file_parameters(self):
+        import anndata
+        try:
+            adata = anndata.read_h5ad(self._file_name)
+            self.n_rows, self.n_cols = adata.shape
+            all_el = self.n_rows * self.n_cols
+            if sp.issparse(adata.X):
+                self.sparsity = (all_el - adata.X.tocsr().count_nonzero()) / all_el
+            else:
+                self.sparsity = (all_el - np.count_nonzero(adata.X)) / all_el
+        except OSError:
+            pass
+
+    def _load_data(self, skip_row=None, skip_col=None, **kwargs):
+        import anndata
+        adata = anndata.read_h5ad(self._file_name)
+
+        if skip_row is not None:
+            mask = np.array([not skip_row(i) for i in range(adata.shape[0])])
+            self._use_rows_mask = mask
+        else:
+            self._use_rows_mask = np.ones(adata.shape[0], dtype=bool)
+        if skip_col is not None:
+            mask = np.array([not skip_col(i) for i in range(adata.shape[1])])
+            self._use_cols_mask = mask
+        else:
+            self._use_cols_mask = np.ones(adata.shape[1], dtype=bool)
+
+        adata = adata[self._use_rows_mask, :]
+        adata = adata[:, self._use_cols_mask]
+
+        attrs = [ContinuousVariable.make(str(g)) for g in adata.var_names]
+        meta_df = pd.DataFrame({key: adata.obs[key] for key in adata.obs.keys()})
+
+        X = adata.X.toarray() if sp.issparse(adata.X) else adata.X
+
         return attrs, X, meta_df, meta_df.index
 
 

setup.py

@@ -35,6 +35,7 @@ def include_documentation(local_dir, install_dir):
             'fastdtw==0.3.2',
             'pandas>=0.23',
             'loompy>=2.0.10',
+            'anndata>=0.6.21',
             'numpy',
             'scikit-learn',
         ],