scMultiChat 0.1.0__py3-none-any.whl

MultiChat/Heterogeneous_g_emb/preprocessing/_qc.py

@@ -0,0 +1,727 @@
+"""Quality Control"""
+
+import numpy as np
+from scipy.sparse import (
+    issparse,
+    csr_matrix,
+)
+import re
+
+
+def cal_qc(adata, expr_cutoff=1):
+    """Calculate quality control metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the feature is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {'rna','atac'},case insensitive
+    Returns
+    -------
+    updates `adata` with the following fields.
+    n_counts: `pandas.Series` (`adata.var['n_counts']`,dtype `int`)
+       The number of read count each gene has.
+    n_cells: `pandas.Series` (`adata.var['n_cells']`,dtype `int`)
+       The number of cells in which each gene is expressed.
+    pct_cells: `pandas.Series` (`adata.var['pct_cells']`,dtype `float`)
+       The percentage of cells in which each gene is expressed.
+    n_counts: `pandas.Series` (`adata.obs['n_counts']`,dtype `int`)
+       The number of read count each cell has.
+    n_genes: `pandas.Series` (`adata.obs['n_genes']`,dtype `int`)
+       The number of genes expressed in each cell.
+    pct_genes: `pandas.Series` (`adata.obs['pct_genes']`,dtype `float`)
+       The percentage of genes expressed in each cell.
+    n_peaks: `pandas.Series` (`adata.obs['n_peaks']`,dtype `int`)
+       The number of peaks expressed in each cell.
+    pct_peaks: `pandas.Series` (`adata.obs['pct_peaks']`,dtype `int`)
+       The percentage of peaks expressed in each cell.
+    pct_mt: `pandas.Series` (`adata.obs['pct_mt']`,dtype `float`)
+       the percentage of counts in mitochondrial genes
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+
+    n_counts = adata.X.sum(axis=0).A1
+    adata.var['n_counts'] = n_counts
+    n_samples = (adata.X >= expr_cutoff).sum(axis=0).A1
+    adata.var['n_samples'] = n_samples
+    adata.var['pct_samples'] = n_samples/adata.shape[0]
+
+    n_counts = adata.X.sum(axis=1).A1
+    adata.obs['n_counts'] = n_counts
+    n_features = (adata.X >= expr_cutoff).sum(axis=1).A1
+    adata.obs['n_features'] = n_features
+    adata.obs['pct_features'] = n_features/adata.shape[1]
+
+
+def cal_qc_rna(adata, expr_cutoff=1):
+    """Calculate quality control metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the feature is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {'rna','atac'},case insensitive
+    Returns
+    -------
+    updates `adata` with the following fields.
+    n_counts: `pandas.Series` (`adata.var['n_counts']`,dtype `int`)
+       The number of read count each gene has.
+    n_cells: `pandas.Series` (`adata.var['n_cells']`,dtype `int`)
+       The number of cells in which each gene is expressed.
+    pct_cells: `pandas.Series` (`adata.var['pct_cells']`,dtype `float`)
+       The percentage of cells in which each gene is expressed.
+    n_counts: `pandas.Series` (`adata.obs['n_counts']`,dtype `int`)
+       The number of read count each cell has.
+    n_genes: `pandas.Series` (`adata.obs['n_genes']`,dtype `int`)
+       The number of genes expressed in each cell.
+    pct_genes: `pandas.Series` (`adata.obs['pct_genes']`,dtype `float`)
+       The percentage of genes expressed in each cell.
+    n_peaks: `pandas.Series` (`adata.obs['n_peaks']`,dtype `int`)
+       The number of peaks expressed in each cell.
+    pct_peaks: `pandas.Series` (`adata.obs['pct_peaks']`,dtype `int`)
+       The percentage of peaks expressed in each cell.
+    pct_mt: `pandas.Series` (`adata.obs['pct_mt']`,dtype `float`)
+       the percentage of counts in mitochondrial genes
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+
+    n_counts = adata.X.sum(axis=0).A1
+    adata.var['n_counts'] = n_counts
+    n_cells = (adata.X >= expr_cutoff).sum(axis=0).A1
+    adata.var['n_cells'] = n_cells
+    adata.var['pct_cells'] = n_cells/adata.shape[0]
+
+    n_counts = adata.X.sum(axis=1).A1
+    adata.obs['n_counts'] = n_counts
+    n_features = (adata.X >= expr_cutoff).sum(axis=1).A1
+    adata.obs['n_genes'] = n_features
+    adata.obs['pct_genes'] = n_features/adata.shape[1]
+    r = re.compile("^MT-", flags=re.IGNORECASE)
+    mt_genes = list(filter(r.match, adata.var_names))
+    if len(mt_genes) > 0:
+        n_counts_mt = adata[:, mt_genes].X.sum(axis=1).A1
+        adata.obs['pct_mt'] = n_counts_mt/n_counts
+    else:
+        adata.obs['pct_mt'] = 0
+
+
+def cal_qc_atac(adata, expr_cutoff=1):
+    """Calculate quality control metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the feature is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {'rna','atac'},case insensitive
+    Returns
+    -------
+    updates `adata` with the following fields.
+    n_counts: `pandas.Series` (`adata.var['n_counts']`,dtype `int`)
+       The number of read count each gene has.
+    n_cells: `pandas.Series` (`adata.var['n_cells']`,dtype `int`)
+       The number of cells in which each gene is expressed.
+    pct_cells: `pandas.Series` (`adata.var['pct_cells']`,dtype `float`)
+       The percentage of cells in which each gene is expressed.
+    n_counts: `pandas.Series` (`adata.obs['n_counts']`,dtype `int`)
+       The number of read count each cell has.
+    n_genes: `pandas.Series` (`adata.obs['n_genes']`,dtype `int`)
+       The number of genes expressed in each cell.
+    pct_genes: `pandas.Series` (`adata.obs['pct_genes']`,dtype `float`)
+       The percentage of genes expressed in each cell.
+    n_peaks: `pandas.Series` (`adata.obs['n_peaks']`,dtype `int`)
+       The number of peaks expressed in each cell.
+    pct_peaks: `pandas.Series` (`adata.obs['pct_peaks']`,dtype `int`)
+       The percentage of peaks expressed in each cell.
+    pct_mt: `pandas.Series` (`adata.obs['pct_mt']`,dtype `float`)
+       the percentage of counts in mitochondrial genes
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+
+    n_counts = adata.X.sum(axis=0).A1
+    adata.var['n_counts'] = n_counts
+    n_cells = (adata.X >= expr_cutoff).sum(axis=0).A1
+    adata.var['n_cells'] = n_cells
+    adata.var['pct_cells'] = n_cells/adata.shape[0]
+
+    n_counts = adata.X.sum(axis=1).A1
+    adata.obs['n_counts'] = n_counts
+    n_features = (adata.X >= expr_cutoff).sum(axis=1).A1
+    adata.obs['n_peaks'] = n_features
+    adata.obs['pct_peaks'] = n_features/adata.shape[1]
+
+
+def filter_samples(adata,
+                   min_n_features=1,
+                   max_n_features=None,
+                   min_pct_features=None,
+                   max_pct_features=None,
+                   min_n_counts=None,
+                   max_n_counts=None,
+                   expr_cutoff=1):
+    """Filter out samples based on different metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    min_n_features: `int`, optional (default: None)
+        Minimum number of features expressed
+    min_pct_features: `float`, optional (default: None)
+        Minimum percentage of features expressed
+    min_n_counts: `int`, optional (default: None)
+        Minimum number of read count for one cell
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the gene is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {{'rna','atac'}},case insensitive
+    Returns
+    -------
+    updates `adata` with a subset of cells that pass the filtering.
+    updates `adata` with the following fields if cal_qc() was not performed.
+    n_counts: `pandas.Series` (`adata.obs['n_counts']`,dtype `int`)
+       The number of read count each cell has.
+    n_genes: `pandas.Series` (`adata.obs['n_genes']`,dtype `int`)
+       The number of genes expressed in each cell.
+    pct_genes: `pandas.Series` (`adata.obs['pct_genes']`,dtype `float`)
+       The percentage of genes expressed in each cell.
+    n_peaks: `pandas.Series` (`adata.obs['n_peaks']`,dtype `int`)
+       The number of peaks expressed in each cell.
+    pct_peaks: `pandas.Series` (`adata.obs['pct_peaks']`,dtype `int`)
+       The percentage of peaks expressed in each cell.
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+    if 'n_counts' in adata.obs_keys():
+        n_counts = adata.obs['n_counts']
+    else:
+        n_counts = np.sum(adata.X, axis=1).A
+        adata.obs['n_counts'] = n_counts
+    if 'n_features' in adata.obs_keys():
+        n_features = adata.obs['n_features']
+    else:
+        n_features = np.sum(adata.X >= expr_cutoff, axis=1).A1
+        adata.obs['n_features'] = n_features
+    if 'pct_features' in adata.obs_keys():
+        pct_features = adata.obs['pct_features']
+    else:
+        pct_features = n_features/adata.shape[1]
+        adata.obs['pct_features'] = pct_features
+
+    print('before filtering: ')
+    print(f"{adata.shape[0]} samples, {adata.shape[1]} feature")
+    if sum(list(map(lambda x: x is None,
+                    [min_n_features,
+                     min_pct_features,
+                     min_n_counts,
+                     max_n_features,
+                     max_pct_features,
+                     max_n_counts]))) == 6:
+        print('No filtering')
+    else:
+        cell_subset = np.ones(len(adata.obs_names), dtype=bool)
+        if min_n_features is not None:
+            print('filter samples based on min_n_features')
+            cell_subset = (n_features >= min_n_features) & cell_subset
+        if max_n_features is not None:
+            print('filter samples based on max_n_features')
+            cell_subset = (n_features <= max_n_features) & cell_subset
+        if min_pct_features is not None:
+            print('filter samples based on min_pct_features')
+            cell_subset = (pct_features >= min_pct_features) & cell_subset
+        if max_pct_features is not None:
+            print('filter samples based on max_pct_features')
+            cell_subset = (pct_features <= max_pct_features) & cell_subset
+        if min_n_counts is not None:
+            print('filter samples based on min_n_counts')
+            cell_subset = (n_counts >= min_n_counts) & cell_subset
+        if max_n_counts is not None:
+            print('filter samples based on max_n_counts')
+            cell_subset = (n_counts <= max_n_counts) & cell_subset
+        adata._inplace_subset_obs(cell_subset)
+        print('after filtering out low-quality samples: ')
+        print(f"{adata.shape[0]} samples, {adata.shape[1]} feature")
+    return None
+
+
+def filter_cells_rna(adata,
+                     min_n_genes=None,
+                     max_n_genes=None,
+                     min_pct_genes=None,
+                     max_pct_genes=None,
+                     min_n_counts=None,
+                     max_n_counts=None,
+                     expr_cutoff=1):
+    """Filter out cells for RNA-seq based on different metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    min_n_genes: `int`, optional (default: None)
+        Minimum number of genes expressed
+    min_pct_genes: `float`, optional (default: None)
+        Minimum percentage of genes expressed
+    min_n_counts: `int`, optional (default: None)
+        Minimum number of read count for one cell
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the gene is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {{'rna','atac'}},case insensitive
+    Returns
+    -------
+    updates `adata` with a subset of cells that pass the filtering.
+    updates `adata` with the following fields if cal_qc() was not performed.
+    n_counts: `pandas.Series` (`adata.obs['n_counts']`,dtype `int`)
+       The number of read count each cell has.
+    n_genes: `pandas.Series` (`adata.obs['n_genes']`,dtype `int`)
+       The number of genes expressed in each cell.
+    pct_genes: `pandas.Series` (`adata.obs['pct_genes']`,dtype `float`)
+       The percentage of genes expressed in each cell.
+    n_peaks: `pandas.Series` (`adata.obs['n_peaks']`,dtype `int`)
+       The number of peaks expressed in each cell.
+    pct_peaks: `pandas.Series` (`adata.obs['pct_peaks']`,dtype `int`)
+       The percentage of peaks expressed in each cell.
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+    if 'n_counts' in adata.obs_keys():
+        n_counts = adata.obs['n_counts']
+    else:
+        n_counts = np.sum(adata.X, axis=1).A1
+        adata.obs['n_counts'] = n_counts
+
+    if 'n_genes' in adata.obs_keys():
+        n_genes = adata.obs['n_genes']
+    else:
+        n_genes = np.sum(adata.X >= expr_cutoff, axis=1).A1
+        adata.obs['n_genes'] = n_genes
+    if 'pct_genes' in adata.obs_keys():
+        pct_genes = adata.obs['pct_genes']
+    else:
+        pct_genes = n_genes/adata.shape[1]
+        adata.obs['pct_genes'] = pct_genes
+
+    print('before filtering: ')
+    print(f"{adata.shape[0]} cells,  {adata.shape[1]} genes")
+    if sum(list(map(lambda x: x is None,
+                    [min_n_genes,
+                     min_pct_genes,
+                     min_n_counts,
+                     max_n_genes,
+                     max_pct_genes,
+                     max_n_counts]))) == 6:
+        print('No filtering')
+    else:
+        cell_subset = np.ones(len(adata.obs_names), dtype=bool)
+        if min_n_genes is not None:
+            print('filter cells based on min_n_genes')
+            cell_subset = (n_genes >= min_n_genes) & cell_subset
+        if max_n_genes is not None:
+            print('filter cells based on max_n_genes')
+            cell_subset = (n_genes <= max_n_genes) & cell_subset
+        if min_pct_genes is not None:
+            print('filter cells based on min_pct_genes')
+            cell_subset = (pct_genes >= min_pct_genes) & cell_subset
+        if max_pct_genes is not None:
+            print('filter cells based on max_pct_genes')
+            cell_subset = (pct_genes <= max_pct_genes) & cell_subset
+        if min_n_counts is not None:
+            print('filter cells based on min_n_counts')
+            cell_subset = (n_counts >= min_n_counts) & cell_subset
+        if max_n_counts is not None:
+            print('filter cells based on max_n_counts')
+            cell_subset = (n_counts <= max_n_counts) & cell_subset
+        adata._inplace_subset_obs(cell_subset)
+        print('after filtering out low-quality cells: ')
+        print(f"{adata.shape[0]} cells,  {adata.shape[1]} genes")
+    return None
+
+
+def filter_cells_atac(adata,
+                      min_n_peaks=None,
+                      max_n_peaks=None,
+                      min_pct_peaks=None,
+                      max_pct_peaks=None,
+                      min_n_counts=None,
+                      max_n_counts=None,
+                      expr_cutoff=1):
+    """Filter out cells for ATAC-seq based on different metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    min_n_peaks: `int`, optional (default: None)
+        Minimum number of peaks expressed
+    min_pct_peaks: `float`, optional (default: None)
+        Minimum percentage of peaks expressed
+    min_n_counts: `int`, optional (default: None)
+        Minimum number of read count for one cell
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the gene is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {{'rna','atac'}},case insensitive
+    Returns
+    -------
+    updates `adata` with a subset of cells that pass the filtering.
+    updates `adata` with the following fields if cal_qc() was not performed.
+    n_counts: `pandas.Series` (`adata.obs['n_counts']`,dtype `int`)
+       The number of read count each cell has.
+    n_genes: `pandas.Series` (`adata.obs['n_genes']`,dtype `int`)
+       The number of genes expressed in each cell.
+    pct_genes: `pandas.Series` (`adata.obs['pct_genes']`,dtype `float`)
+       The percentage of genes expressed in each cell.
+    n_peaks: `pandas.Series` (`adata.obs['n_peaks']`,dtype `int`)
+       The number of peaks expressed in each cell.
+    pct_peaks: `pandas.Series` (`adata.obs['pct_peaks']`,dtype `int`)
+       The percentage of peaks expressed in each cell.
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+    if 'n_counts' in adata.obs_keys():
+        n_counts = adata.obs['n_counts']
+    else:
+        n_counts = np.sum(adata.X, axis=1).A1
+        adata.obs['n_counts'] = n_counts
+
+    if 'n_peaks' in adata.obs_keys():
+        n_peaks = adata.obs['n_peaks']
+    else:
+        n_peaks = np.sum(adata.X >= expr_cutoff, axis=1).A1
+        adata.obs['n_peaks'] = n_peaks
+    if 'pct_peaks' in adata.obs_keys():
+        pct_peaks = adata.obs['pct_peaks']
+    else:
+        pct_peaks = n_peaks/adata.shape[1]
+        adata.obs['pct_peaks'] = pct_peaks
+
+    print('before filtering: ')
+    print(f"{adata.shape[0]} cells,  {adata.shape[1]} peaks")
+    if sum(list(map(lambda x: x is None,
+                    [min_n_peaks,
+                     min_pct_peaks,
+                     min_n_counts,
+                     max_n_peaks,
+                     max_pct_peaks,
+                     max_n_counts]))) == 6:
+        print('No filtering')
+    else:
+        cell_subset = np.ones(len(adata.obs_names), dtype=bool)
+        if min_n_peaks is not None:
+            print('filter cells based on min_n_peaks')
+            cell_subset = (n_peaks >= min_n_peaks) & cell_subset
+        if max_n_peaks is not None:
+            print('filter cells based on max_n_peaks')
+            cell_subset = (n_peaks <= max_n_peaks) & cell_subset
+        if min_pct_peaks is not None:
+            print('filter cells based on min_pct_peaks')
+            cell_subset = (pct_peaks >= min_pct_peaks) & cell_subset
+        if max_pct_peaks is not None:
+            print('filter cells based on max_pct_peaks')
+            cell_subset = (pct_peaks <= max_pct_peaks) & cell_subset
+        if min_n_counts is not None:
+            print('filter cells based on min_n_counts')
+            cell_subset = (n_counts >= min_n_counts) & cell_subset
+        if max_n_counts is not None:
+            print('filter cells based on max_n_counts')
+            cell_subset = (n_counts <= max_n_counts) & cell_subset
+        adata._inplace_subset_obs(cell_subset)
+        print('after filtering out low-quality cells: ')
+        print(f"{adata.shape[0]} cells,  {adata.shape[1]} peaks")
+    return None
+
+
+def filter_genes(adata,
+                 min_n_cells=3,
+                 max_n_cells=None,
+                 min_pct_cells=None,
+                 max_pct_cells=None,
+                 min_n_counts=None,
+                 max_n_counts=None,
+                 expr_cutoff=1):
+    """Filter out features based on different metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    min_n_cells: `int`, optional (default: 5)
+        Minimum number of cells expressing one feature
+    min_pct_cells: `float`, optional (default: None)
+        Minimum percentage of cells expressing one feature
+    min_n_counts: `int`, optional (default: None)
+        Minimum number of read count for one feature
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the feature is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {{'rna','atac'}},case insensitive
+
+    Returns
+    -------
+    updates `adata` with a subset of features that pass the filtering.
+    updates `adata` with the following fields if cal_qc() was not performed.
+    n_counts: `pandas.Series` (`adata.var['n_counts']`,dtype `int`)
+       The number of read count each gene has.
+    n_cells: `pandas.Series` (`adata.var['n_cells']`,dtype `int`)
+       The number of cells in which each gene is expressed.
+    pct_cells: `pandas.Series` (`adata.var['pct_cells']`,dtype `float`)
+       The percentage of cells in which each gene is expressed.
+    """
+
+    feature = 'genes'
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+
+    if 'n_counts' in adata.var_keys():
+        n_counts = adata.var['n_counts']
+    else:
+        n_counts = np.sum(adata.X, axis=0).A1
+        adata.var['n_counts'] = n_counts
+    if 'n_cells' in adata.var_keys():
+        n_cells = adata.var['n_cells']
+    else:
+        n_cells = np.sum(adata.X >= expr_cutoff, axis=0).A1
+        adata.var['n_cells'] = n_cells
+    if 'pct_cells' in adata.var_keys():
+        pct_cells = adata.var['pct_cells']
+    else:
+        pct_cells = n_cells/adata.shape[0]
+        adata.var['pct_cells'] = pct_cells
+
+    print('Before filtering: ')
+    print(str(adata.shape[0])+' cells, ' + str(adata.shape[1])+' '+feature)
+    if sum(list(map(lambda x: x is None,
+                    [min_n_cells, min_pct_cells, min_n_counts,
+                     max_n_cells, max_pct_cells, max_n_counts,
+                     ]))) == 6:
+        print('No filtering')
+    else:
+        feature_subset = np.ones(len(adata.var_names), dtype=bool)
+        if min_n_cells is not None:
+            print('Filter '+feature+' based on min_n_cells')
+            feature_subset = (n_cells >= min_n_cells) & feature_subset
+        if max_n_cells is not None:
+            print('Filter '+feature+' based on max_n_cells')
+            feature_subset = (n_cells <= max_n_cells) & feature_subset
+        if min_pct_cells is not None:
+            print('Filter '+feature+' based on min_pct_cells')
+            feature_subset = (pct_cells >= min_pct_cells) & feature_subset
+        if max_pct_cells is not None:
+            print('Filter '+feature+' based on max_pct_cells')
+            feature_subset = (pct_cells <= max_pct_cells) & feature_subset
+        if min_n_counts is not None:
+            print('Filter '+feature+' based on min_n_counts')
+            feature_subset = (n_counts >= min_n_counts) & feature_subset
+        if max_n_counts is not None:
+            print('Filter '+feature+' based on max_n_counts')
+            feature_subset = (n_counts <= max_n_counts) & feature_subset
+        adata._inplace_subset_var(feature_subset)
+        print('After filtering out low-expressed '+feature+': ')
+        print(str(adata.shape[0])+' cells, ' + str(adata.shape[1])+' '+feature)
+    return None
+
+
+def filter_peaks(adata,
+                 min_n_cells=5,
+                 max_n_cells=None,
+                 min_pct_cells=None,
+                 max_pct_cells=None,
+                 min_n_counts=None,
+                 max_n_counts=None,
+                 expr_cutoff=1):
+    """Filter out features based on different metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    min_n_cells: `int`, optional (default: 5)
+        Minimum number of cells expressing one feature
+    min_pct_cells: `float`, optional (default: None)
+        Minimum percentage of cells expressing one feature
+    min_n_counts: `int`, optional (default: None)
+        Minimum number of read count for one feature
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the feature is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {{'rna','atac'}},case insensitive
+
+    Returns
+    -------
+    updates `adata` with a subset of features that pass the filtering.
+    updates `adata` with the following fields if cal_qc() was not performed.
+    n_counts: `pandas.Series` (`adata.var['n_counts']`,dtype `int`)
+       The number of read count each gene has.
+    n_cells: `pandas.Series` (`adata.var['n_cells']`,dtype `int`)
+       The number of cells in which each gene is expressed.
+    pct_cells: `pandas.Series` (`adata.var['pct_cells']`,dtype `float`)
+       The percentage of cells in which each gene is expressed.
+    """
+
+    feature = 'peaks'
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+
+    if 'n_counts' in adata.var_keys():
+        n_counts = adata.var['n_counts']
+    else:
+        n_counts = np.sum(adata.X, axis=0).A1
+        adata.var['n_counts'] = n_counts
+    if 'n_cells' in adata.var_keys():
+        n_cells = adata.var['n_cells']
+    else:
+        n_cells = np.sum(adata.X >= expr_cutoff, axis=0).A1
+        adata.var['n_cells'] = n_cells
+    if 'pct_cells' in adata.var_keys():
+        pct_cells = adata.var['pct_cells']
+    else:
+        pct_cells = n_cells/adata.shape[0]
+        adata.var['pct_cells'] = pct_cells
+
+    print('Before filtering: ')
+    print(str(adata.shape[0])+' cells, ' + str(adata.shape[1])+' '+feature)
+    if sum(list(map(lambda x: x is None,
+                    [min_n_cells, min_pct_cells, min_n_counts,
+                     max_n_cells, max_pct_cells, max_n_counts,
+                     ]))) == 6:
+        print('No filtering')
+    else:
+        feature_subset = np.ones(len(adata.var_names), dtype=bool)
+        if min_n_cells is not None:
+            print('Filter '+feature+' based on min_n_cells')
+            feature_subset = (n_cells >= min_n_cells) & feature_subset
+        if max_n_cells is not None:
+            print('Filter '+feature+' based on max_n_cells')
+            feature_subset = (n_cells <= max_n_cells) & feature_subset
+        if min_pct_cells is not None:
+            print('Filter '+feature+' based on min_pct_cells')
+            feature_subset = (pct_cells >= min_pct_cells) & feature_subset
+        if max_pct_cells is not None:
+            print('Filter '+feature+' based on max_pct_cells')
+            feature_subset = (pct_cells <= max_pct_cells) & feature_subset
+        if min_n_counts is not None:
+            print('Filter '+feature+' based on min_n_counts')
+            feature_subset = (n_counts >= min_n_counts) & feature_subset
+        if max_n_counts is not None:
+            print('Filter '+feature+' based on max_n_counts')
+            feature_subset = (n_counts <= max_n_counts) & feature_subset
+        adata._inplace_subset_var(feature_subset)
+        print('After filtering out low-expressed '+feature+': ')
+        print(str(adata.shape[0])+' cells, ' + str(adata.shape[1])+' '+feature)
+    return None
+
+
+def filter_features(adata,
+                    min_n_samples=5,
+                    max_n_samples=None,
+                    min_pct_samples=None,
+                    max_pct_samples=None,
+                    min_n_counts=None,
+                    max_n_counts=None,
+                    expr_cutoff=1):
+    """Filter out features based on different metrics.
+
+    Parameters
+    ----------
+    adata: AnnData
+        Annotated data matrix.
+    min_n_cells: `int`, optional (default: 5)
+        Minimum number of cells expressing one feature
+    min_pct_cells: `float`, optional (default: None)
+        Minimum percentage of cells expressing one feature
+    min_n_counts: `int`, optional (default: None)
+        Minimum number of read count for one feature
+    expr_cutoff: `float`, optional (default: 1)
+        Expression cutoff.
+        If greater than expr_cutoff,the feature is considered 'expressed'
+    assay: `str`, optional (default: 'rna')
+            Choose from {{'rna','atac'}},case insensitive
+
+    Returns
+    -------
+    updates `adata` with a subset of features that pass the filtering.
+    updates `adata` with the following fields if cal_qc() was not performed.
+    n_counts: `pandas.Series` (`adata.var['n_counts']`,dtype `int`)
+       The number of read count each gene has.
+    n_cells: `pandas.Series` (`adata.var['n_cells']`,dtype `int`)
+       The number of cells in which each gene is expressed.
+    pct_cells: `pandas.Series` (`adata.var['pct_cells']`,dtype `float`)
+       The percentage of cells in which each gene is expressed.
+    """
+
+    if not issparse(adata.X):
+        adata.X = csr_matrix(adata.X)
+    if 'n_counts' in adata.var_keys():
+        n_counts = adata.var['n_counts']
+    else:
+        n_counts = np.sum(adata.X, axis=0).A1
+        adata.var['n_counts'] = n_counts
+    if 'n_samples' in adata.var_keys():
+        n_samples = adata.var['n_samples']
+    else:
+        n_samples = np.sum(adata.X >= expr_cutoff, axis=0).A1
+        adata.var['n_samples'] = n_samples
+    if 'pct_samples' in adata.var_keys():
+        pct_samples = adata.var['pct_samples']
+    else:
+        pct_samples = n_samples/adata.shape[0]
+        adata.var['pct_samples'] = pct_samples
+
+    print('Before filtering: ')
+    print(f"{adata.shape[0]} samples,  {adata.shape[1]} features")
+
+    if sum(list(map(lambda x: x is None,
+                    [min_n_samples, min_pct_samples, min_n_counts,
+                     max_n_samples, max_pct_samples, max_n_counts,
+                     ]))) == 6:
+        print('No filtering')
+    else:
+        feature_subset = np.ones(len(adata.var_names), dtype=bool)
+        if min_n_samples is not None:
+            print('Filter features based on min_n_samples')
+            feature_subset = (n_samples >= min_n_samples) & feature_subset
+        if max_n_samples is not None:
+            print('Filter features based on max_n_samples')
+            feature_subset = (n_samples <= max_n_samples) & feature_subset
+        if min_pct_samples is not None:
+            print('Filter features based on min_pct_samples')
+            feature_subset = (pct_samples >= min_pct_samples) & feature_subset
+        if max_pct_samples is not None:
+            print('Filter features based on max_pct_samples')
+            feature_subset = (pct_samples <= max_pct_samples) & feature_subset
+        if min_n_counts is not None:
+            print('Filter features based on min_n_counts')
+            feature_subset = (n_counts >= min_n_counts) & feature_subset
+        if max_n_counts is not None:
+            print('Filter features based on max_n_counts')
+            feature_subset = (n_counts <= max_n_counts) & feature_subset
+        adata._inplace_subset_var(feature_subset)
+        print('After filtering out low-expressed features: ')
+        print(f"{adata.shape[0]} samples,  {adata.shape[1]} features")
+    return None