def new_model(self, emission_dists = None, dirichlet_strength = 10.0):
model = self.background_model_creator(self.order, self.num_background_mosaics)
# have we been given any emission dist?
if None == emission_dists:
# no - were we initialised with some distribution?
if None == self.emission_dists:
# no - create a random one
emission_dists = [ hmm.dirichlet_draw(numpy.ones(4)*dirichlet_strength) for k in xrange(self.K) ]
else:
# yes - use it
emission_dists = self.emission_dists
# add the states for the positive strand orientation
positive_states = [
self.model_builder.add_order_0_parameterised_state(
model,
pi=self.p_binding_site,
emission_dist=dist)
for dist in emission_dists
]
# add the states for the negative strand orientation
negative_states = [
self.model_builder.add_order_0_rev_comp_state(
model,
positive_state,
pi=self.p_binding_site)
for positive_state in positive_states
]
negative_states.reverse()
# connect the background to the both first states with equal prob
param = model.add_parameter(self.p_binding_site/2)
for bg in xrange(self.num_background_mosaics):
model.states[bg].add_successor(positive_states[0], param)
model.states[bg].add_successor(negative_states[0], param)
# connect the states in the pssm together
one_param = model.add_parameter(1.0)
for k in xrange(self.K-1):
positive_states[k].add_successor(positive_states[k+1], one_param)
negative_states[k].add_successor(negative_states[k+1], one_param)
# connect the last states back to the background
one_param = model.add_parameter(1.0/self.num_background_mosaics)
for bg in xrange(self.num_background_mosaics):
positive_states[-1].add_successor(model.states[bg], one_param)
negative_states[-1].add_successor(model.states[bg], one_param)
return model
def emission_dist_including_n_mer(self, n_mer, strength=9.0, offset=0, dirichlet_strength = 10.0):
'''
Get an emission dist that incorporates the n-mer
Draws a distribution from a dirichlet of given strength (default 10) then adds the n-mer in the middle
(default # copies of n-mer = 9)
'''
emission_dists = [ hmm.dirichlet_draw(numpy.ones(4)*dirichlet_strength) for k in xrange(self.K_prime) ]
n = len(n_mer)
offset = (self.K - n)/2
for i, base in enumerate(n_mer):
idx = 2*(i+offset)
emission_dists[idx].setflags(write=True)
emission_dists[idx][base] += strength
emission_dists[idx] /= emission_dists[idx].sum()
return emission_dists
def create_mosaic_model(
num_mosaics,
p_transition,
alphabet_size,
order,
dirichlet_prior_strength=None
):
"""
Create a mosaic model.
Each mosaic has an independent parameter that specifies the probability of transitioning to any other
given mosaic (this effectively ties these transition probabilities together).
@arg num_mosaics: The number of mosaics.
@arg p_transition: The probability of leaving a mosaic.
@arg alphabet_size: The size of the output alphabet.
@arg order: The Markov order of the output.
@arg dirichlet_prior_strength: The strength of the uniform prior on the emission probabilities. If None
then a uniform distribution is used.
"""
builder = hmm.pssm.ModelBuilder(order, alphabet_size=alphabet_size)
model = builder.new_model_by_states()
for n in xrange(num_mosaics):
# add the state
if None == dirichlet_prior_strength:
emission_dist = numpy.ones(model.M) / alphabet_size
else:
emission_dist = hmm.dirichlet_draw(numpy.ones(model.M)*dirichlet_prior_strength)
state = builder.add_fully_parameterised_state(
model,
pi=1./num_mosaics,
emission_dist=emission_dist
)
if 1 == num_mosaics:
model.states[0].add_successor(model.states[0], model.add_parameter(1.))
else:
for s1 in model.states:
transition_param = model.add_parameter(p_transition)
no_transition_param = model.add_parameter(1.0 - p_transition)
for s2 in model.states:
s1.add_successor(s2, s1 == s2 and no_transition_param or transition_param)
return model
def create_background_mosaic_model(self, num_mosaics, p_transition, dirichlet_prior_strength):
"""
Create a mosaic model
"""
model = hmm.ModelByStates(self.M, self.order)
transition_param = model.add_parameter(p_transition)
no_transition_param = model.add_parameter(1.0 - p_transition)
for i in xrange(num_mosaics):
self.add_fully_parameterised_state(
model,
emission_dist = hmm.dirichlet_draw(numpy.ones(self.M)*dirichlet_prior_strength)
)
for state_1 in model.states:
for state_2 in model.states:
if state_1 == state_2: state_1.add_successor(state_2, no_transition_param)
else: state_1.add_successor(state_2, transition_param)
return model
def create_mosaic_model(num_mosaics,
p_transition,
alphabet_size,
order,
dirichlet_prior_strength=None):
"""
Create a mosaic model.
Each mosaic has an independent parameter that specifies the probability of transitioning to any other
given mosaic (this effectively ties these transition probabilities together).
@arg num_mosaics: The number of mosaics.
@arg p_transition: The probability of leaving a mosaic.
@arg alphabet_size: The size of the output alphabet.
@arg order: The Markov order of the output.
@arg dirichlet_prior_strength: The strength of the uniform prior on the emission probabilities. If None
then a uniform distribution is used.
"""
builder = hmm.pssm.ModelBuilder(order, alphabet_size=alphabet_size)
model = builder.new_model_by_states()
for n in xrange(num_mosaics):
# add the state
if None == dirichlet_prior_strength:
emission_dist = numpy.ones(model.M) / alphabet_size
else:
emission_dist = hmm.dirichlet_draw(
numpy.ones(model.M) * dirichlet_prior_strength)
state = builder.add_fully_parameterised_state(
model, pi=1. / num_mosaics, emission_dist=emission_dist)
if 1 == num_mosaics:
model.states[0].add_successor(model.states[0], model.add_parameter(1.))
else:
for s1 in model.states:
transition_param = model.add_parameter(p_transition)
no_transition_param = model.add_parameter(1.0 - p_transition)
for s2 in model.states:
s1.add_successor(
s2, s1 == s2 and no_transition_param or transition_param)
return model
def new_model(self, emission_dists = None, dirichlet_strength = 10.0, generate_uniform=False):
model = self.background_model_creator(self.order, self.num_background_mosaics)
# are we generating a uniform emission model?
if generate_uniform:
emission_dists = [ .25 * numpy.ones(4) for k in xrange(self.K_prime) ]
else:
# have we been given any emission dist?
if None == emission_dists:
# no - were we initialised with some distribution?
if None == self.emission_dists:
# no - create a random one
emission_dists = [ hmm.dirichlet_draw(numpy.ones(4)*dirichlet_strength) for k in xrange(self.K_prime) ]
else:
# yes - use it
emission_dists = self.emission_dists
if len(emission_dists) != self.K_prime:
raise RuntimeError('Wrong number of emissions')
# add the states for the positive strand orientation
positive_states = [
self.model_builder.add_order_0_parameterised_state(
model,
pi=self.p_binding_site,
emission_dist=dist)
for dist in emission_dists
]
# add the states for the negative strand orientation
negative_states = [
self.model_builder.add_order_0_rev_comp_state(
model,
positive_state,
pi=self.p_binding_site)
for positive_state in positive_states
]
negative_states.reverse()
# connect the background to the both first states with equal prob
param = model.add_parameter(self.p_binding_site/2)
for bg in xrange(self.num_background_mosaics):
model.states[bg].add_successor(positive_states[0], param)
model.states[bg].add_successor(negative_states[0], param)
# connect the states in the pssm together
one_param = model.add_parameter(1.0)
gap_params = [ model.add_parameter(self.p_gap) for i in xrange(self.K-1) ]
non_gap_params = [ model.add_parameter(1.0-self.p_gap) for i in xrange(self.K-1) ]
for k in xrange(self.K-1):
positive_states[2*k+1].add_successor(positive_states[2*k+2], one_param)
negative_states[2*k+1].add_successor(negative_states[2*k+2], one_param)
positive_states[2*k].add_successor(positive_states[2*k+1], gap_params[k])
negative_states[2*k].add_successor(negative_states[2*k+1], gap_params[self.K-2-k])
positive_states[2*k].add_successor(positive_states[2*k+2], non_gap_params[k])
negative_states[2*k].add_successor(negative_states[2*k+2], non_gap_params[self.K-2-k])
# connect the last states back to the background
one_param = model.add_parameter(1.0/self.num_background_mosaics)
for bg in xrange(self.num_background_mosaics):
positive_states[-1].add_successor(model.states[bg], one_param)
negative_states[-1].add_successor(model.states[bg], one_param)
return model
print "Seeding the random number generator with %d" % seed
# seed all the RNGs that we use
hmm.seed_rng(seed)
numpy.random.seed(seed)
# create something to build the gapped pssms
builder = single_gap.SingleGappedPssmBuilder(K=K, gap_position=K / 2, markov_order=0, M=4)
# create our emission distributions
dirichlet_prior_strengths = [0.01, 0.1, 1.0]
emissions = [
numpy.array([hmm.dirichlet_draw(numpy.ones(builder.M) * strength) for k in xrange(builder.K)])
for strength in dirichlet_prior_strengths
]
gap_emissions = [hmm.dirichlet_draw(numpy.ones(builder.M) * strength) for strength in dirichlet_prior_strengths]
# create out single gapped pssms
pssms = [builder.create(p_gap, non_gap, gap) for non_gap, gap in zip(emissions, gap_emissions)]
# create our complete models (by adding a background model)
p_binding_site = exp_sites_per_sequence / L
models = [
hmm.as_model(
single_gap.add_to_simple_background_model(
model=pssm[0], in_states=pssm[1], out_states=pssm[2], p_binding_site=p_binding_site
complete_model.states[0].add_successor(
complete_model.states[1 + in_state], binding_site_transition_param)
for out_state in out_states:
complete_model.states[1 + out_state].add_successor(
complete_model.states[0], back_to_bg_transition_param)
complete_model.states[0].pi = complete_model.add_parameter(1.)
return complete_model
if '__main__' == __name__:
import numpy
# build a single gapped pssm with some random emissions
builder = SingleGappedPssmBuilder(K=6, gap_index=1, markov_order=0, M=4)
emissions = numpy.array([
hmm.dirichlet_draw(numpy.ones(builder.M) * .1)
for k in xrange(builder.K)
])
emissions[builder.gap_index] = hmm.dirichlet_draw(
numpy.ones(builder.M) * .3)
model_by_states, in_states, out_states = builder.create(
p_gap=.6, emissions=emissions)
# create a background model and add the single gapped pssm to it
complete_model = add_to_simple_background_model(model_by_states,
in_states,
out_states,
p_binding_site=.01)
# convert to other type of model
model = hmm.as_model(complete_model)
back_to_bg_transition_param
)
complete_model.states[0].pi = complete_model.add_parameter(1.)
return complete_model
if '__main__' == __name__:
import numpy
# build a single gapped pssm with some random emissions
builder = SingleGappedPssmBuilder(K=6, gap_index=1, markov_order=0, M=4)
emissions = numpy.array(
[
hmm.dirichlet_draw(numpy.ones(builder.M) * .1)
for k in xrange(builder.K)
]
)
emissions[builder.gap_index] = hmm.dirichlet_draw(numpy.ones(builder.M) * .3)
model_by_states, in_states, out_states = builder.create(
p_gap=.6,
emissions=emissions
)
# create a background model and add the single gapped pssm to it
complete_model = add_to_simple_background_model(
model_by_states,
in_states,
out_states,
p_binding_site=.01)