def sample_cfdna_allele(p, tot_counts, bbdisp):
"""
Sample reads at site from a beta-binomial distribution, where the
probability of sampling the A allele is based on the proportion of the A
allele in the maternal-fetal mixture drawn from a betabinomial distribution
with dispersion parameter bbdisp. If the beta-binomial parameter is set to
np.inf, then the beta-binomial distribution collapses to the binomial.
Args:
p (float): proportion of A allele in maternal-fetal mix
tot_counts (int): the number of reads at the site (depth).
bbdisp (float): beta-binomial dispersion parameter in allele counts
Returns:
The number of A allele reads.
Note:
If bbdisp == np.inf the beta-binomial distribution collapses to the
simple binomial.
Raises:
AssertionError: If p < 0 or > 1
AssertionError: If bbdisp < 1
"""
assert bbdisp >= 1, "Value of bbdisp must be > 1"
assert 0 <= p <= 1, "Value of p must be > 0 and < 1"
#At probs 0 or 1 the beta parameters are undefined, therefore default to
#sampling from a binomial distribution with p= 0 or 1. Always apply the
#sampling variance from the most abundant allele.
flip = False
if p != 0 and p != 1 and bbdisp != np.inf:
if p < 0.5:
p = 1-p
flip = True
a = bbdisp
b = (bbdisp / p) - bbdisp
if flip:
return tot_counts - pymc.rbetabin(a, b, tot_counts)
else:
return pymc.rbetabin(a, b, tot_counts)
else:
return np.random.binomial(tot_counts, p)
def pred(pi=pi, alpha=alpha, beta=beta):
return mc.rbetabin(alpha, beta, n)
def p_pred(pi=pi, delta=delta, n=n_nonzero):
return mc.rbetabin(alpha=pi[~i_zero] * delta[~i_zero] * 50,
beta=(1 - pi[~i_zero]) * delta[~i_zero] * 50,
n=n[~i_zero]) / pl.array(n + 1.e-9, dtype=float)
def pred(pi=pi, alpha=alpha, beta=beta):
return mc.rbetabin(alpha, beta, n)