def analyse(events, subset=None, strata=None):
if strata is None:
bg = estimateBackground(events.data)
else:
bg = estimateBackgroundStratified(events.data, strata)
return fdr.mutex(events[subset], bg[subset])
def pairwise_discover_test(x, g=None, alternative="less", correct=True):
"""
Perform many pairwise mutual exclusivity or co-occurrence tests.
Parameters
----------
x : DiscoverMatrix
g : array_like, optional
An optional grouping vector for the rows of `x`. Pairs of rows within
the same group are not tested.
alternative : {'less', 'greater'}, optional
If 'less', a mutual-exclusivity analysis is performed, if 'greater' a
co-occurrence analysis.
correct : bool, optional
If True, multiple testing correction is performed.
Returns
-------
result : PairwiseDiscoverResult
An object containing the test results for all pairwise combinations.
"""
assert alternative in ["less", "greater"]
events = x.events
bg = x.bg
if g is None:
pFlat, qFlat, pi0 = fdr.mutex(events, bg, alternative == "less")
p = numpy.empty((x.shape[0], ) * 2)
p[:] = numpy.nan
p[numpy.triu_indices_from(p, 1)] = pFlat
q = numpy.empty((x.shape[0], ) * 2)
q[:] = numpy.nan
q[numpy.triu_indices_from(p, 1)] = qFlat
else:
i = numpy.argsort(g)
levels, inverse = numpy.unique(g, return_inverse=True)
blockSizes = numpy.bincount(inverse)
p, q, pi0 = fdr.analyseblockstructure(events[i], bg[i],
alternative == "less",
blockSizes)
j = numpy.argsort(i)
p = p[j[:, numpy.newaxis], j]
q = q[j[:, numpy.newaxis], j]
p = pandas.DataFrame(p, index=x.rownames, columns=x.rownames)
q = pandas.DataFrame(q, index=x.rownames, columns=x.rownames)
return PairwiseDiscoverResult(p, q, pi0, alternative)
def pairwise_discover_test(x, g=None, alternative="less", fdr_method="DBH"):
"""
Perform many pairwise mutual exclusivity or co-occurrence tests.
Parameters
----------
x : DiscoverMatrix
g : array_like, optional
An optional grouping vector for the rows of `x`. Pairs of rows within
the same group are not tested.
alternative : {'less', 'greater'}, optional
If 'less', a mutual-exclusivity analysis is performed, if 'greater' a
co-occurrence analysis.
fdr_method : {'DBH', 'BH'}, optional
The false discovery rate procedure used for multiple testing correction.
If 'DBH', a Benjamini-Hochberg procedure adapted for discrete test statistics
is used. If 'BH', the standard Benjamini-Hochberg procedure is used. The latter
is much faster, but also more conservative than the discrete version.
Returns
-------
result : PairwiseDiscoverResult
An object containing the test results for all pairwise combinations.
"""
assert alternative in ["less", "greater"]
assert fdr_method in ["DBH", "BH"]
discrete_fdr = fdr_method == "DBH"
events = x.events
bg = x.bg
if g is None:
pFlat, qFlat, pi0 = fdr.mutex(events, bg, alternative == "less",
discrete_fdr)
if fdr_method == "BH":
qFlat = false_discovery_rate(pFlat)
pi0 = 1.0
p = numpy.empty((x.shape[0], ) * 2)
p[:] = numpy.nan
p[numpy.triu_indices_from(p, 1)] = pFlat
q = numpy.empty((x.shape[0], ) * 2)
q[:] = numpy.nan
q[numpy.triu_indices_from(p, 1)] = qFlat
else:
i = numpy.argsort(g)
levels, inverse = numpy.unique(g, return_inverse=True)
blockSizes = numpy.bincount(inverse)
p, q, pi0 = fdr.analyseblockstructure(events[i], bg[i],
alternative == "less",
blockSizes, discrete_fdr)
if fdr_method == "BH":
q = false_discovery_rate(p)
pi0 = 1.0
j = numpy.argsort(i)
p = p[j[:, numpy.newaxis], j]
q = q[j[:, numpy.newaxis], j]
p = pandas.DataFrame(p, index=x.rownames, columns=x.rownames)
q = pandas.DataFrame(q, index=x.rownames, columns=x.rownames)
return PairwiseDiscoverResult(p, q, pi0, alternative, fdr_method)