def _calc_reference_confidence(self, n_ref, n_total):
"""Performs the calculation described in reference_confidence()."""
if n_ref < 0:
raise ValueError('n_ref={} must be >= 0'.format(n_ref))
if n_total < n_ref:
raise ValueError('n_total={} must be >= n_ref={}'.format(
n_total, n_ref))
if self.options.ploidy != 2:
raise ValueError('ploidy={} but we only support ploidy=2'.format(
self.options.ploidy))
if n_total == 0:
# No coverage case - all likelihoods are log10 of 1/3, 1/3, 1/3.
log10_probs = genomics_math.normalize_log10_probs(
[-1.0, -1.0, -1.0])
else:
n_alts = n_total - n_ref
log10_p_ref = genomics_math.log10_binomial(n_alts, n_total,
self.options.p_error)
log10_p_het = genomics_math.log10_binomial(
n_ref, n_total, 1.0 / self.options.ploidy)
log10_p_hom_alt = genomics_math.log10_binomial(
n_ref, n_total, self.options.p_error)
log10_probs = genomics_math.normalize_log10_probs(
[log10_p_ref, log10_p_het, log10_p_hom_alt])
gq = genomics_math.log10_ptrue_to_phred(log10_probs[0],
self.options.max_gq)
gq = int(min(np.floor(gq), self.options.max_gq))
return gq, log10_probs
def _calc_reference_confidence(self, n_ref, n_total):
"""Performs the calculation described in reference_confidence()."""
if n_ref < 0:
raise ValueError('n_ref={} must be >= 0'.format(n_ref))
if n_total < n_ref:
raise ValueError('n_total={} must be >= n_ref={}'.format(n_total, n_ref))
if self.options.ploidy != 2:
raise ValueError('ploidy={} but we only support ploidy=2'.format(
self.options.ploidy))
if n_total == 0:
# No coverage case - all likelihoods are log10 of 1/3, 1/3, 1/3.
log10_probs = genomics_math.normalize_log10_probs([-1.0, -1.0, -1.0])
else:
n_alts = n_total - n_ref
log10_p_ref = genomics_math.log10_binomial(n_alts, n_total,
self.options.p_error)
log10_p_het = genomics_math.log10_binomial(n_ref, n_total,
1.0 / self.options.ploidy)
log10_p_hom_alt = genomics_math.log10_binomial(n_ref, n_total,
self.options.p_error)
log10_probs = genomics_math.normalize_log10_probs(
[log10_p_ref, log10_p_het, log10_p_hom_alt])
gq = genomics_math.log10_ptrue_to_phred(log10_probs[0], self.options.max_gq)
gq = int(min(np.floor(gq), self.options.max_gq))
return gq, log10_probs
def scaled_likelihoods(self):
"""Returns the scaled likelihood of each genotype."""
if not all(bool(x) for x in self._genotype_likelihood_containers):
raise ValueError(
'All genotypes must have some probability mass: {}'.format(
self._genotype_likelihood_containers))
return genomics_math.normalize_log10_probs([
genomics_math.log10sumexp(unscaled)
for unscaled in self._genotype_likelihood_containers
])
def test_normalize_log10_probs(self, log10_probs, expected):
npt.assert_allclose(
genomics_math.normalize_log10_probs(log10_probs), expected, atol=1e-6)
def test_normalize_log10_probs(self, log10_probs, expected):
npt.assert_allclose(
genomics_math.normalize_log10_probs(log10_probs), expected, atol=1e-6)