def downsample_counts(adata,
target_counts=20000,
random_state=0,
replace=True,
copy=False):
"""Downsample counts so that each cell has no more than `target_counts`.
Cells with fewer counts than `target_counts` are unaffected by this. This
has been implemented by M. D. Luecken.
Parameters
----------
adata : :class:`~anndata.AnnData`
Annotated data matrix.
target_counts : `int` (default: 20,000)
Target number of counts for downsampling. Cells with more counts than
'target_counts' will be downsampled to have 'target_counts' counts.
random_state : `int` or `None`, optional (default: 0)
Random seed to change subsampling.
replace : `bool`, optional (default: `True`)
Whether to sample the counts with replacement.
copy : `bool`, optional (default: `False`)
If an :class:`~anndata.AnnData` is passed, determines whether a copy
is returned.
Returns
-------
AnnData, None
Depending on `copy` returns or updates an `adata` with downsampled `.X`.
"""
if copy:
adata = adata.copy()
adata.X = adata.X.astype(np.integer) # Numba doesn't want floats
if issparse(adata.X):
X = adata.X
if not isspmatrix_csr(X):
X = csr_matrix(X)
totals = np.ravel(X.sum(axis=1))
under_target = np.nonzero(totals > target_counts)[0]
cols = np.split(X.data.view(), X.indptr[1:-1])
for colidx in under_target:
col = cols[colidx]
downsample_cell(col,
target_counts,
random_state=random_state,
replace=replace,
inplace=True)
if not isspmatrix_csr(adata.X): # Put it back
adata.X = type(adata.X)(X)
else:
totals = np.ravel(adata.X.sum(axis=1))
under_target = np.nonzero(totals > target_counts)[0]
adata.X[under_target, :] = \
np.apply_along_axis(downsample_cell, 1, adata.X[under_target, :],
target_counts, random_state=random_state, replace=replace)
if copy: return adata
def filter_genes_fano_deprecated(X, Ecutoff, Vcutoff):
"""Filter genes by fano factor and mean.
See `filter_genes_dispersion`.
Reference: Weinreb et al. (2017).
"""
if issparse(X):
raise ValueError('Not defined for sparse input. See `filter_genes_dispersion`.')
mean_filter = np.mean(X, axis=0) > Ecutoff
var_filter = np.var(X, axis=0) / (np.mean(X, axis=0) + .0001) > Vcutoff
gene_subset = np.nonzero(np.all([mean_filter, var_filter], axis=0))[0]
return gene_subset
def filter_genes_cv_deprecated(X, Ecutoff, cvFilter):
"""Filter genes by coefficient of variance and mean.
See `filter_genes_dispersion`.
Reference: Weinreb et al. (2017).
"""
if issparse(X):
raise ValueError('Not defined for sparse input. See `filter_genes_dispersion`.')
mean_filter = np.mean(X, axis=0) > Ecutoff
var_filter = np.std(X, axis=0) / (np.mean(X, axis=0) + .0001) > cvFilter
gene_subset = np.nonzero(np.all([mean_filter, var_filter], axis=0))[0]
return gene_subset