def get_pre_Q(cls,
log_counts, codon_distn,
ts, tv, syn, nonsyn, compo, asym_compo,
natural_theta,
):
cls.check_theta(theta)
kappa = natural_theta[0]
omega = natural_theta[1]
nt_distn = markovutil.ratios_to_distn(theta[2:5])
codon_distn = codon1994.get_f1x4_codon_distn(compo, nt_distn)
pre_Q = codon1994.get_pre_Q(
ts, tv, syn, nonsyn,
codon_distn, kappa, omega)
return pre_Q
def get_neg_ll(cls,
patterns, pattern_weights,
stationary_distn,
ts, tv, syn, nonsyn,
theta,
):
"""
This model has only a single omega parameter.
@param theta: vector of free variables with sensitivities
"""
# unpack theta
log_mus = theta[0:3]
log_kappa = theta[3]
log_omega = theta[4]
# construct the transition matrices
transition_matrices = []
for i in range(3):
mu = algopy.exp(log_mus[i])
kappa = algopy.exp(log_kappa)
omega = algopy.exp(log_omega)
pre_Q = codon1994.get_pre_Q(
ts, tv, syn, nonsyn,
stationary_distn,
kappa, omega)
Q = markovutil.pre_Q_to_Q(pre_Q, stationary_distn, mu)
P = algopy.expm(Q)
transition_matrices.append(P)
# return the neg log likelihood
ov = range(4)
v_to_children = {3 : [0, 1, 2]}
de_to_P = {
(3, 0) : transition_matrices[0],
(3, 1) : transition_matrices[1],
(3, 2) : transition_matrices[2],
}
root_prior = stationary_distn
log_likelihood = alignll.fast_fels(
#log_likelihood = alignll.fels(
ov, v_to_children, de_to_P, root_prior,
patterns, pattern_weights,
)
neg_ll = -log_likelihood
print neg_ll
return neg_ll