def test_params(my_N, my_ind, my_v):
result = get_xlmhg_test_result(my_N, my_ind, X=1)
assert isinstance(result, mHGResult)
result = get_xlmhg_test_result(my_N, my_ind, L=my_N)
assert isinstance(result, mHGResult)
result = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.05)
assert isinstance(result, mHGResult)
table = np.empty((my_N + 1, my_N + 1), np.longdouble)
result = get_xlmhg_test_result(my_N, my_ind, table=table)
assert isinstance(result, mHGResult)
def test_figure(tmpdir):
v = np.uint8([1,0,1,1,0,1] + [0]*12 + [1,0])
X = 3
L = 10
N = v.size
indices = np.uint16(np.nonzero(v)[0])
result = xlmhg.get_xlmhg_test_result(N, indices, X=X, L=L)
fig = xlmhg.get_result_figure(result)
output_file = text(tmpdir.join('plot1.html'))
plot(fig, filename=output_file, auto_open=False)
assert os.path.isfile(output_file)
fig = xlmhg.get_result_figure(result, width=500, height=350)
output_file = text(tmpdir.join('plot2.html'))
plot(fig, filename=output_file, auto_open=False)
assert os.path.isfile(output_file)
fig = xlmhg.get_result_figure(result, show_title=True, show_inset=False)
output_file = text(tmpdir.join('plot3.html'))
plot(fig, filename=output_file, auto_open=False)
assert os.path.isfile(output_file)
def test_significant2(my_N, my_ind):
"""Test if we return the exact p-value for a significant test when
requested, even if we don't need to calculate it in order to determine that
the test is significant."""
res = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.045,
exact_pval='if_significant')
assert res.pval == 0.0244453044375645
def test_limit_pval(my_incredible_pval_v):
"""Test differential accuracy of PVAL1 and PVAL2."""
# PVAL1 algorithm should handle this without problems
N = my_incredible_pval_v.size
ind = np.uint16(np.nonzero(my_incredible_pval_v)[0])
res = get_xlmhg_test_result(N, ind)
assert res.stat == 1.5112233509292993e-216
assert res.cutoff == 200
assert res.pval == res.stat
res = get_xlmhg_test_result(N, ind, use_alg1=True)
# PVAL2 algorithm should report an invalid p-value
# (either <= 0 or unrealistically large; in this case < 0)
# and the front-end should replace that with the O(1)-bound
assert res.stat == 1.5112233509292993e-216
assert res.cutoff == 200
assert res.stat < res.pval < 1e-200
def my_rank_based_result(my_v, my_ranked_genes, my_gene_set):
N = my_v.size
X = 1
L = N
indices = np.uint16(np.nonzero(my_v)[0])
ind_genes = [my_ranked_genes[i] for i in indices]
assert set(ind_genes) == my_gene_set.genes
## stat, n_star, pval = xlmhg_test(my_v, X, L)
res = get_xlmhg_test_result(N, indices, X, L)
# result = GSEResult(stat, n_star, pval, N, X, L, my_gene_set,
# sel, sel_genes)
result = RankBasedGSEResult(my_gene_set, N, indices, ind_genes, X, L, res.stat, res.cutoff, res.pval)
return result
def my_rank_based_result(my_v, my_ranked_genes, my_gene_set):
N = my_v.size
X = 1
L = N
indices = np.uint16(np.nonzero(my_v)[0])
ind_genes = [my_ranked_genes[i] for i in indices]
assert set(ind_genes) == my_gene_set.genes
## stat, n_star, pval = xlmhg_test(my_v, X, L)
res = get_xlmhg_test_result(N, indices, X, L)
#result = GSEResult(stat, n_star, pval, N, X, L, my_gene_set,
# sel, sel_genes)
result = RankBasedGSEResult(my_gene_set, N, indices, ind_genes, X, L,
res.stat, res.cutoff, res.pval)
return result
def my_rank_based_result(my_matrix, my_v):
indices = np.uint16(np.nonzero(my_v)[0])
ind_genes = [my_matrix.genes[i] for i in indices]
gs_genes = list(ind_genes)
gene_set = GeneSet(genes=gs_genes, id='Random1', name='Random gene Set 1')
N = my_v.size
X = 1
L = N
## stat, n_star, pval = xlmhg_test(my_v, X, L)
res = get_xlmhg_test_result(N, indices, X, L)
result = RankBasedGSEResult(gene_set, N, indices, ind_genes, X, L,
res.stat, res.cutoff, res.pval)
return result
def test_figure_double_axis(tmpdir):
v = np.uint8([1,0,1,1,0,1] + [0]*12 + [1,0])
X = 3
L = 10
N = v.size
indices = np.uint16(np.nonzero(v)[0])
result = xlmhg.get_xlmhg_test_result(N, indices, X=X, L=L)
fig = xlmhg.get_result_figure(result, plot_fold_enrichment=True)
output_file = text(tmpdir.join('plot4.html'))
plot(fig, filename=output_file, auto_open=False)
assert os.path.isfile(output_file)
def test_significant1(my_N, my_ind):
"""Test if we return the exact p-value for a significant test when
requested."""
res = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.07,
exact_pval='if_significant')
assert res.pval == 0.0244453044375645
def test_ONbound(my_N, my_ind):
"""Test if we return the O(N)-bound instead of the exact
p-value, if the bound is equal to or smaller than `pval_thresh`."""
res = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.045,
exact_pval='if_necessary')
assert res.pval == 0.04179566563467492
def test_O1bound(my_N, my_ind):
"""Test if we return the O(1)-bound if that's "<=" `pval_thresh`"""
res = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.07,
exact_pval='if_necessary')
assert res.pval == 0.0696594427244582
def test_alg1(my_N, my_ind):
"""Test if we can use PVAL1 to calculate p-value."""
res = get_xlmhg_test_result(my_N, my_ind, use_alg1=True)
assert res.stat == 0.01393188854489164
assert res.cutoff == 6
assert res.pval == 0.0244453044375645
def test_result(my_ind, my_v):
N = my_v.size
result = get_xlmhg_test_result(N, my_ind)
assert isinstance(result, mHGResult)
def test_lowerbound(my_N, my_ind):
"""Test if we return the O(1)-bound when stat > pval_thresh."""
res = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.01,
exact_pval='if_necessary')
assert res.pval == 0.0696594427244582
def get_rank_based_enrichment(
self, ranked_genes, pval_thresh=0.05,
X_frac=0.25, X_min=5, L=None,
adjust_pval_thresh=True, escore_pval_thresh=None,
exact_pval='always', gene_set_ids=None, table=None):
"""Test for gene set enrichment at the top of a ranked list of genes.
This function uses the XL-mHG test to identify enriched gene sets.
This function also calculates XL-mHG E-scores for the enriched gene
sets, using ``escore_pval_thresh`` as the p-value threshold "psi".
Parameters
----------
ranked_genes : list of str
The ranked list of genes.
pval_thresh : float, optional
The p-value threshold used to determine significance.
See also ``adjust_pval_thresh``. [0.05]
X_frac : float, optional
The min. fraction of genes from a gene set required for enrichment. [0.25]
X_min : int, optional
The min. no. of genes from a gene set required for enrichment. [5]
L : int, optional
The lowest cutoff to test for enrichment. If ``None``,
int(0.25*(no. of genes)) will be used. [None]
adjust_pval_thresh : bool, optional
Whether to adjust the p-value thershold for multiple testing,
using the Bonferroni method. [True]
escore_pval_thresh : float or None, optional
The "psi" p-value threshold used in calculating E-scores. [None]
exact_pval : str
Choices are: "always", "if_significant", "if_necessary". Parameter
will be passed to `xlmhg.get_xlmhg_test_result`. ["always"]
gene_set_ids : list of str or None, optional
A list of gene set IDs to specify which gene sets should be tested for enrichment. If ``None``, all gene sets will be tested. [None]
table : 2-dim numpy.ndarray of type numpy.longdouble or None, optional
The dynamic programming table used by the algorithm for calculating XL-mHG p-values. Passing this avoids memory re-allocation when calling this function repetitively. [None]
Returns
-------
list of `RankBasedGSEResult`
A list of all significantly enriched gene sets.
"""
if isinstance(X_frac, (int, np.integer)):
X_frac = float(X_frac)
# type checks
assert isinstance(ranked_genes, Iterable)
assert isinstance(pval_thresh, (float, np.float))
assert isinstance(X_frac, (float, np.float))
assert isinstance(X_min, (int, np.integer))
if L is not None:
assert isinstance(L, (int, np.integer))
assert isinstance(adjust_pval_thresh, bool)
assert isinstance(exact_pval, (str, _oldstr))
if escore_pval_thresh is not None:
assert isinstance(escore_pval_thresh, (float, np.float))
if gene_set_ids is not None:
assert isinstance(gene_set_ids, Iterable)
if table is not None:
assert isinstance(table, np.ndarray) and \
np.issubdtype(table.dtype, np.longdouble)
if L is None:
L = int(len(ranked_genes)/4.0)
gene_set_coll = self._gene_set_coll
gene_memberships = self._gene_memberships
# postpone this
if escore_pval_thresh is None:
# if no separate E-score p-value threshold is specified, use the
# p-value threshold (this results in conservative E-scores)
logger.warning('No E-score p-value threshold supplied. '
'The E-score p-value threshold will be set to the'
'global significance threshold. This will result '
'in conservative E-scores.')
# test only some terms?
if gene_set_ids is not None:
gs_indices = np.int64([self._gene_set_coll.index(id_)
for id_ in gene_set_ids])
gene_sets = [gene_set_coll[id_] for id_ in gene_set_ids]
gene_set_coll = GeneSetCollection(gene_sets)
gene_memberships = gene_memberships[:, gs_indices] # not a view!
# reorder rows in annotation matrix to match the given gene ranking
# also exclude genes not in the ranking
unknown = 0
L_adj = L
sel = []
filtered_genes = []
logger.debug('Looking up indices for %d genes...' % len(ranked_genes))
for i, g in enumerate(ranked_genes):
assert isinstance(g, (str, _oldstr))
try:
idx = self._genome.index(g)
except ValueError:
unknown += 1
# adjust L if the gene was above the original L cutoff
if i < L:
L_adj -= 1
else:
sel.append(idx)
filtered_genes.append(g)
sel = np.int64(sel)
logger.debug('Adjusted L: %d', L_adj)
# the following also copies the data (not a view)
gene_memberships = gene_memberships[sel, :]
N, m = gene_memberships.shape
if unknown > 0:
# Some genes in the ranked list were unknown (i.e., not present in
# the specified genome).
logger.warn('%d / %d unknown genes (%.1f %%), will be ignored.',
unknown, len(ranked_genes),
100 * (unknown / float(len(ranked_genes))))
# Determine the number of gene set genes above the L'th cutoff,
# for all gene sets. This quantity is useful, because we don't need
# to perform any more work for gene sets that have less than X genes
# above the cutoff.
k_above_L = np.sum(gene_memberships[:L_adj, :], axis=0, dtype=np.int64)
# Determine the number of genes below the L'th cutoff, for all gene
# sets.
k_below_L = np.sum(gene_memberships[L_adj:, :], axis=0, dtype=np.int64)
# Determine the total number K of genes in each gene set that are
# present in the ranked list (this is equal to k_above_L + k_below_L)
K_vec = k_above_L + k_below_L
# Determine the largest K across all gene sets.
K_max = np.amax(K_vec)
# Determine X for all gene sets.
X = np.amax(
np.c_[np.tile(X_min, m), np.int64(np.ceil(X_frac * K_vec))],
axis=1)
# Determine the number of tests (we do not conduct a test if the
# total number of gene set genes in the ranked list is below X).
num_tests = np.sum(K_vec-X >= 0)
logger.info('Conducting %d tests.', num_tests)
# determine Bonferroni-corrected p-value, if desired
final_pval_thresh = pval_thresh
if adjust_pval_thresh and num_tests > 0:
final_pval_thresh /= float(num_tests)
logger.info('Using Bonferroni-corrected p-value threshold: %.1e',
final_pval_thresh)
if escore_pval_thresh is None:
escore_pval_thresh = final_pval_thresh
elif escore_pval_thresh < final_pval_thresh:
logger.warning('The E-score p-value threshold is smaller than '
'the p-value threshold. Setting E-score p-value '
'threshold to the p-value threshold.')
escore_pval_thresh = final_pval_thresh
# Prepare the matrix that holds the dynamic programming table for
# the calculation of the XL-mHG p-value.
if table is None:
table = np.empty((K_max+1, N+1), dtype=np.longdouble)
else:
if table.shape[0] < K_max+1 or table.shape[1] < N+1:
raise ValueError(
'The supplied array is too small (%d x %d) to hold the '
'entire dynamic programming table. The required size is'
'%d x %d (rows x columns).'
% (table.shape[0], table.shape[1], K_max+1, N+1))
# find enriched GO terms
# logger.info('Testing %d gene sets for enrichment...', m)
logger.debug('(N=%d, X_frac=%.2f, X_min=%d, L=%d; K_max=%d)',
len(ranked_genes), X_frac, X_min, L, K_max)
enriched = []
num_tests = 0 # number of tests conducted
for j in range(m):
# determine gene set-specific value for X
X = max(X_min, int(ceil(X_frac * float(K_vec[j]))))
# Determine significance of gene set enrichment using the XL-mHG
# test (only if there are at least X gene set genes in the list).
if K_vec[j] >= X:
num_tests += 1
# We only need to perform the XL-mHG test if there are enough
# gene set genes above the L'th cutoff (otherwise, pval = 1.0).
if k_above_L[j] >= X:
# perform test
# Determine the ranks of the gene set genes in the
# ranked list.
indices = np.uint16(np.nonzero(gene_memberships[:, j])[0])
res = xlmhg.get_xlmhg_test_result(
N, indices, X, L, pval_thresh=final_pval_thresh,
escore_pval_thresh=escore_pval_thresh,
exact_pval=exact_pval, table=table)
# check if gene set is significantly enriched
if res.pval <= final_pval_thresh:
# generate RankedGSEResult
ind_genes = [ranked_genes[i] for i in indices]
gse_result = RankBasedGSEResult(
gene_set_coll[j], N, indices, ind_genes,
X, L, res.stat, res.cutoff, res.pval,
escore_pval_thresh=escore_pval_thresh
)
enriched.append(gse_result)
# report results
q = len(enriched)
ignored = m - num_tests
if ignored > 0:
logger.debug('%d / %d gene sets (%.1f%%) had less than X genes '
'annotated with them and were ignored.',
ignored, m, 100 * (ignored / float(m)))
logger.info('%d / %d gene sets were found to be significantly '
'enriched (p-value <= %.1e).', q, m, final_pval_thresh)
return enriched
def batch_xlmhg(marker_exp, c_list, coi, X=None, L=None):
"""Applies XL-mHG test to a gene expression matrix, gene by gene.
Outputs a 3-column DataFrame representing statistical results of XL-mHG.
:param marker_exp: A DataFrame whose rows are cell identifiers, columns are
gene identifiers, and values are float values representing gene
expression.
:param c_list: A Series whose indices are cell identifiers, and whose
values are the cluster which that cell is part of.
:param coi: The cluster of interest.
:param X: An integer to be used as argument to the XL-mHG test.
:param L: An integer to be used as argument to the XL-mHG test.
:returns: A matrix with arbitrary row indices, whose columns are the gene
name, stat, cutoff, and pval outputs of the XL-mHG test; of
float, int, and float type respectively. Their names are 'gene',
'mHG_stat', 'mHG_cutoff', and 'mHG_pval'.
:rtype: pandas.DataFrame
"""
# * 1 converts to integer
mem_list = (c_list == coi) * 1
#count_n = 0
#count_2n = 0
#Count the number of cells in the cluster, store into count_n
count_n = np.sum(mem_list)
'''
for cell in mem_list:
if cell == 1:
count_n = count_n + 1
else:
continue
'''
#Twice the number of cells in the cluster
#count_2n = count_n * 2
#Set X and L params
if X is None:
X = np.int(.15 * count_n)
else:
X = np.int(X * count_n)
if L is None:
if 2 * count_n >= marker_exp.shape[0]:
L = np.int(marker_exp.shape[0])
else:
L = np.int(2 * count_n)
#L = marker_exp.shape[0]
print('X = ' + str(X))
print('L = ' + str(L))
print('Cluster size ' + str(count_n))
xlmhg = marker_exp.apply(lambda col: hg.xlmhg_test(mem_list.reindex(
col.sort_values(ascending=False).index).values,
X=X,
L=L))
xlmhg_1 = marker_exp.apply(lambda col: hg.get_xlmhg_test_result(
N=len(mem_list),
indices=np.array(np.where(
np.array(
mem_list.reindex(col.sort_values(ascending=False).index).values
== 1))[0],
dtype='int64').astype('uint16'),
X=X,
L=L,
pval_thresh=1e-12,
escore_pval_thresh=1e-1,
tol=1e-12).escore)
output = pd.DataFrame()
output['gene_1'] = xlmhg.index
output[['mHG_stat', 'mHG_cutoff', 'mHG_pval'
]] = pd.DataFrame(xlmhg.values.tolist(),
columns=['mHG_stat', 'mHG_cutoff', 'mHG_pval'])
'''
print(xlmhg_1)
output['escore'] = np.array(xlmhg_1,dtype=float)
output.fillna(0,inplace=True)
print(output.sort_values(by='escore', ascending=False))
time.sleep(1000)
'''
return output
def test_pval_necessary(my_N, my_ind):
"""Test if we return the p-value when it is necessary."""
res = get_xlmhg_test_result(my_N, my_ind, pval_thresh=0.04,
exact_pval='if_necessary')
assert res.pval == 0.0244453044375645
def test_non_contiguous(my_N, my_ind):
"""Test if non-contiguous arrays result in a ValueError."""
with pytest.raises(ValueError):
result = get_xlmhg_test_result(my_N, my_ind[::-1])
def test_table_too_small(my_N, my_ind):
"""Test if ValueError is raised when the supplied array is too small."""
K = my_ind.size
with pytest.raises(ValueError):
table = np.empty(((my_N-K), (my_N-K)), np.longdouble)
result = get_xlmhg_test_result(my_N, my_ind, table=table)
def get_rank_based_enrichment(
self,
ranked_genes: List[str],
pval_thresh: float = 0.05,
X_frac: float = 0.25,
X_min: int = 5,
L: int = None,
adjust_pval_thresh: bool = True,
escore_pval_thresh: float = None,
exact_pval: str = 'always',
gene_set_ids: List[str] = None,
table: np.ndarray = None) -> RankBasedGSEResult:
"""Test for gene set enrichment at the top of a ranked list of genes.
This function uses the XL-mHG test to identify enriched gene sets.
This function also calculates XL-mHG E-scores for the enriched gene
sets, using ``escore_pval_thresh`` as the p-value threshold "psi".
Parameters
----------
ranked_gene_ids : list of str
The ranked list of gene IDs.
pval_thresh : float, optional
The p-value threshold used to determine significance.
See also ``adjust_pval_thresh``. [0.05]
X_frac : float, optional
The min. fraction of genes from a gene set required for enrichment. [0.25]
X_min : int, optional
The min. no. of genes from a gene set required for enrichment. [5]
L : int, optional
The lowest cutoff to test for enrichment. If ``None``,
int(0.25*(no. of genes)) will be used. [None]
adjust_pval_thresh : bool, optional
Whether to adjust the p-value thershold for multiple testing,
using the Bonferroni method. [True]
escore_pval_thresh : float or None, optional
The "psi" p-value threshold used in calculating E-scores. If
``None``, will be set to p-value threshold. [None]
exact_pval : str
Choices are: "always", "if_significant", "if_necessary". Parameter
will be passed to `xlmhg.get_xlmhg_test_result`. ["always"]
gene_set_ids : list of str or None, optional
A list of gene set IDs to specify which gene sets should be tested for enrichment. If ``None``, all gene sets will be tested. [None]
table : 2-dim numpy.ndarray of type numpy.longdouble or None, optional
The dynamic programming table used by the algorithm for calculating XL-mHG p-values. Passing this avoids memory re-allocation when calling this function repetitively. [None]
Returns
-------
list of `RankBasedGSEResult`
A list of all significantly enriched gene sets.
"""
# make sure X_frac is a float (e.g., if specified as 0)
X_frac = float(X_frac)
if table is not None:
if not np.issubdtype(table.dtype, np.longdouble):
raise TypeError('The provided array for storing the dynamic '
'programming table must be of type '
'"longdouble"!')
if L is None:
L = int(len(ranked_genes) / 4.0)
gene_set_coll = self._gene_set_coll
gene_memberships = self._gene_memberships
# postpone this
if escore_pval_thresh is None:
# if no separate E-score p-value threshold is specified, use the
# p-value threshold (this results in conservative E-scores)
logger.warning('No E-score p-value threshold supplied. '
'The E-score p-value threshold will be set to the'
'global significance threshold. This will result '
'in conservative E-scores.')
# test only some terms?
if gene_set_ids is not None:
gs_indices = np.int64(
[self._gene_set_coll.index(id_) for id_ in gene_set_ids])
gene_sets = [gene_set_coll[id_] for id_ in gene_set_ids]
gene_set_coll = GeneSetCollection(gene_sets)
gene_memberships = gene_memberships[:, gs_indices] # not a view!
# reorder rows in annotation matrix to match the given gene ranking
# also exclude genes not in the ranking
unknown = 0
L_adj = L
sel = []
filtered_genes = []
logger.debug('Looking up indices for %d genes...' % len(ranked_genes))
for i, g in enumerate(ranked_genes):
try:
idx = self._gene_indices[g]
except KeyError:
unknown += 1
# adjust L if the gene was above the original L cutoff
if i < L:
L_adj -= 1
else:
sel.append(idx)
filtered_genes.append(g)
sel = np.int64(sel)
logger.debug('Adjusted L: %d', L_adj)
# the following also copies the data (not a view)
gene_memberships = gene_memberships[sel, :]
N, m = gene_memberships.shape
if unknown > 0:
# Some genes in the ranked list were unknown (i.e., not present in
# the specified genome).
logger.warn('%d / %d unknown genes (%.1f %%), will be ignored.',
unknown, len(ranked_genes),
100 * (unknown / float(len(ranked_genes))))
# Determine the number of gene set genes above the L'th cutoff,
# for all gene sets. This quantity is useful, because we don't need
# to perform any more work for gene sets that have less than X genes
# above the cutoff.
k_above_L = np.sum(gene_memberships[:L_adj, :], axis=0, dtype=np.int64)
# Determine the number of genes below the L'th cutoff, for all gene
# sets.
k_below_L = np.sum(gene_memberships[L_adj:, :], axis=0, dtype=np.int64)
# Determine the total number K of genes in each gene set that are
# present in the ranked list (this is equal to k_above_L + k_below_L)
K_vec = k_above_L + k_below_L
# Determine the largest K across all gene sets.
K_max = np.amax(K_vec)
# Determine X for all gene sets.
X = np.amax(np.c_[np.tile(X_min, m),
np.int64(np.ceil(X_frac * K_vec))],
axis=1)
# Determine the number of tests (we do not conduct a test if the
# total number of gene set genes in the ranked list is below X).
num_tests = np.sum(K_vec - X >= 0)
logger.info('Conducting %d tests.', num_tests)
# determine Bonferroni-corrected p-value, if desired
final_pval_thresh = pval_thresh
if adjust_pval_thresh and num_tests > 0:
final_pval_thresh /= float(num_tests)
logger.info('Using Bonferroni-corrected p-value threshold: %.1e',
final_pval_thresh)
if escore_pval_thresh is None:
escore_pval_thresh = final_pval_thresh
elif escore_pval_thresh < final_pval_thresh:
logger.warning('The E-score p-value threshold is smaller than '
'the p-value threshold. Setting E-score p-value '
'threshold to the p-value threshold.')
escore_pval_thresh = final_pval_thresh
# Prepare the matrix that holds the dynamic programming table for
# the calculation of the XL-mHG p-value.
if table is None:
table = np.empty((K_max + 1, N + 1), dtype=np.longdouble)
else:
if table.shape[0] < K_max + 1 or table.shape[1] < N + 1:
raise ValueError(
'The supplied array is too small (%d x %d) to hold the '
'entire dynamic programming table. The required size is'
'%d x %d (rows x columns).' %
(table.shape[0], table.shape[1], K_max + 1, N + 1))
# find enriched GO terms
# logger.info('Testing %d gene sets for enrichment...', m)
logger.debug('(N=%d, X_frac=%.2f, X_min=%d, L=%d; K_max=%d)',
len(ranked_genes), X_frac, X_min, L, K_max)
enriched = []
num_tests = 0 # number of tests conducted
for j in range(m):
# determine gene set-specific value for X
X = max(X_min, int(ceil(X_frac * float(K_vec[j]))))
# Determine significance of gene set enrichment using the XL-mHG
# test (only if there are at least X gene set genes in the list).
if K_vec[j] >= X:
num_tests += 1
# We only need to perform the XL-mHG test if there are enough
# gene set genes above the L'th cutoff (otherwise, pval = 1.0).
if k_above_L[j] >= X:
# perform test
# Determine the ranks of the gene set genes in the
# ranked list.
indices = np.uint16(np.nonzero(gene_memberships[:, j])[0])
res = xlmhg.get_xlmhg_test_result(
N,
indices,
X,
L,
pval_thresh=final_pval_thresh,
escore_pval_thresh=escore_pval_thresh,
exact_pval=exact_pval,
table=table)
# check if gene set is significantly enriched
if res.pval <= final_pval_thresh:
# generate RankedGSEResult
ind_genes = [ranked_genes[i] for i in indices]
gse_result = RankBasedGSEResult(
gene_set_coll[j],
N,
indices,
ind_genes,
X,
L,
res.stat,
res.cutoff,
res.pval,
escore_pval_thresh=escore_pval_thresh)
enriched.append(gse_result)
# report results
q = len(enriched)
ignored = m - num_tests
if ignored > 0:
logger.debug(
'%d / %d gene sets (%.1f%%) had less than X genes '
'annotated with them and were ignored.', ignored, m,
100 * (ignored / float(m)))
logger.info(
'%d / %d gene sets were found to be significantly '
'enriched (p-value <= %.1e).', q, m, final_pval_thresh)
return enriched
