def run_sampler_on_event(gene, ni, ne, nb, read_len, overhang_len, num_iters,
output_dir, confidence_level=.95):
"""
Run sampler on a two-isoform gene event.
"""
print "Running sampler on a two-isoform event..."
print " - Gene label: ", gene.label, gene
print " - NI, NE, NB: %d, %d, %d" %(ni, ne, nb)
print "Using default sampler parameters."
if gene.chrom != None:
# Index output by chromosome
print "Indexing by chromosome..."
output_dir = os.path.join(output_dir, gene.chrom)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
output_filename = os.path.join(output_dir, gene.label)
samples = []
cred_interval = []
num_isoforms = len(gene.isoforms)
burn_in = 500
lag = 10
hyperparameters = ones(num_isoforms)
proposal_diag = 0.05
sigma = set_diag(zeros([num_isoforms-1, num_isoforms-1]),
proposal_diag)
sampler_params = {'read_len': read_len,
'overhang_len': overhang_len,
'uniform_proposal': False,
'sigma_proposal': sigma}
sampler = MISOSampler(sampler_params, log_dir=output_dir)
reads = read_counts_to_read_list(ni, ne, nb)
t1 = time.time()
sampler_results = sampler.run_sampler(num_iters, reads, gene, hyperparameters,
sampler_params, output_filename, burn_in=burn_in,
lag=lag)
if not sampler_results:
return (samples, cred_interval)
samples = sampler_results[1]
# Compute credible intervals
cred_interval = ht.compute_credible_intervals(samples, confidence_level=confidence_level)
t2 = time.time()
print " - Sampler run took %s seconds." %(str(t2-t1))
# return samples and credible intervals
return (samples, cred_interval)
def plot_two_iso_samples(self,
fig=None,
isoform_index=0,
num_rows=1,
num_cols=1,
subplot_start=1,
plots_dir=None,
map_estimate=None,
simulation_num=1,
plot_intervals=False,
value_to_label=None,
label=None,
plot_filename=None,
bins=None,
bbox_coords=None,
with_legend=True,
title=None,
vanilla=False,
plot_mean=False,
normed=False,
fig_dims=(5, 5)):
"""
Plot a set of samples for Psi of a two isoform gene.
"""
if not fig:
sampled_psi_fig = plt.figure(figsize=fig_dims, dpi=300)
else:
sampled_psi_fig = fig
ax = sampled_psi_fig.add_subplot(num_rows, num_cols, subplot_start)
num_iters = int(self.params['iters'])
burn_in = int(self.params['burn_in'])
lag = int(self.params['lag'])
percent_acceptance = float(self.params['percent_accept'])
proposal_type = self.params['proposal_type']
plt.rcParams['font.size'] = 10
show_spines(ax, ['left', 'bottom'])
bins = bins
assert((value_to_label == None and label == None) or \
(value_to_label != None and label != None))
# retrieve samples
samples_to_plot = self.samples[:, isoform_index]
# picasso blue #0276FD
if not vanilla:
if bins != None:
plt.hist(samples_to_plot,
align='mid',
lw=0.5,
facecolor='#0276FD',
edgecolor='#ffffff')
else:
plt.hist(samples_to_plot,
align='mid',
lw=0.5,
facecolor='#0276FD',
edgecolor='#ffffff')
else:
plt.hist(samples_to_plot,
align='mid',
facecolor='#0276FD',
edgecolor='#0276FD')
plt.xlabel(r'${\hat{\Psi}}_{\mathregular{MISO}}$')
plt.ylabel('Frequency')
plt.xlim([0, 1])
# Normalize samples
if normed:
yticks = list(plt.gca().get_yticks())
print "yticks: ", yticks
ytick_labels = [
"%.2f" % (float(ytick) / float(normed)) for ytick in yticks
]
ax.set_yticklabels(ytick_labels)
# samples_to_plot = samples_to_plot / float(len(samples_to_plot))
# curr_tick_labels = [label.get_label() for label in ax.get_yticklabels()]
# print "Current tick labels: ", curr_tick_labels
# new_tick_labels = []
# for label in curr_tick_labels:
# if len(label) > 0:
# new_label = "%.1f" %(float(label) / normed)
# else:
# new_label = ""
# new_tick_labels.append(new_label)
# #ax.set_yticklabels(new_tick_labels)
curr_axes = plt.gca()
# Plot MAP estimate for same data
if map_estimate:
l = plt.axvline(x=map_estimate,
color='b',
linewidth=1.2,
ls='-',
label=r'${\hat{\Psi}}_{MAP}\ =\ %.2f$' %
(map_estimate))
# Plot true Psi
if self.true_psi:
plot_id = "%dsimul_%diters_%dburnin_%dlag_%s_truepsi_%.2f.pdf" \
%(simulation_num, num_iters, burn_in, lag, proposal_type, self.true_psi)
l = plt.axvline(x=self.true_psi,
color='r',
linewidth=1.2,
ls='-',
label=r'True $\Psi$')
else:
# Unknown true Psi
plot_id = "%dsimul_%diters_%dburnin_%dlag_%s_%s_truepsi.pdf" \
%(simulation_num, num_iters, burn_in, lag, proposal_type, 'unknown')
if value_to_label:
l = plt.axvline(x=value_to_label,
color='r',
linewidth=1.2,
ls='-',
label=label)
# plot credible intervals if given
if plot_intervals:
# print "Plotting %.2f confidence intervals" %(plot_intervals * 100)
interval_c1, interval_c2 = ht.compute_credible_intervals(
samples_to_plot, plot_intervals)
plt.axvline(x=interval_c1,
color='#999999',
linewidth=0.7,
ls='--',
label=r'%d' % (plot_intervals * 100) + '% CI')
plt.axvline(x=interval_c2, color='#999999', linewidth=0.7, ls='--')
if plot_mean:
sample_mean = mean(samples_to_plot)
plt.axvline(x=sample_mean, color='r', linewidth=0.8, label='Mean')
if with_legend and (plot_intervals or self.true_psi):
if not bbox_coords:
lg = plt.legend(handletextpad=0.172,
borderpad=0.01,
labelspacing=.008,
handlelength=1.4,
loc='best',
numpoints=1)
else:
lg = plt.legend(handletextpad=0.172,
borderpad=0.01,
labelspacing=.008,
handlelength=1.4,
loc='best',
numpoints=1,
bbox_to_anchor=bbox_coords)
lg.get_frame().set_linewidth(0)
for t in lg.get_texts():
t.set_fontsize(8)
if title:
plt.title(title)
if plots_dir:
if not plot_filename:
plt.savefig(plots_dir + "sampled_psi_hist_%s" % (plot_id))
else:
plt.savefig(plots_dir + plot_filename + '.pdf')
return curr_axes
def run_sampler_on_event(gene,
ni,
ne,
nb,
read_len,
overhang_len,
num_iters,
output_dir,
confidence_level=.95):
"""
Run sampler on a two-isoform gene event.
"""
print "Running sampler on a two-isoform event..."
print " - Gene label: ", gene.label, gene
print " - NI, NE, NB: %d, %d, %d" % (ni, ne, nb)
print "Using default sampler parameters."
if gene.chrom != None:
# Index output by chromosome
print "Indexing by chromosome..."
output_dir = os.path.join(output_dir, gene.chrom)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
output_filename = os.path.join(output_dir, gene.label)
samples = []
cred_interval = []
num_isoforms = len(gene.isoforms)
burn_in = 500
lag = 10
hyperparameters = ones(num_isoforms)
proposal_diag = 0.05
sigma = set_diag(zeros([num_isoforms - 1, num_isoforms - 1]),
proposal_diag)
sampler_params = {
'read_len': read_len,
'overhang_len': overhang_len,
'uniform_proposal': False,
'sigma_proposal': sigma
}
sampler = MISOSampler(sampler_params, log_dir=output_dir)
reads = read_counts_to_read_list(ni, ne, nb)
t1 = time.time()
sampler_results = sampler.run_sampler(num_iters,
reads,
gene,
hyperparameters,
sampler_params,
output_filename,
burn_in=burn_in,
lag=lag)
if not sampler_results:
return (samples, cred_interval)
samples = sampler_results[1]
# Compute credible intervals
cred_interval = ht.compute_credible_intervals(
samples, confidence_level=confidence_level)
t2 = time.time()
print " - Sampler run took %s seconds." % (str(t2 - t1))
# return samples and credible intervals
return (samples, cred_interval)
def plot_two_iso_samples(self, fig=None, isoform_index=0, num_rows=1, num_cols=1, subplot_start=1,
plots_dir=None, map_estimate=None, simulation_num=1,
plot_intervals=False, value_to_label=None, label=None, plot_filename=None,
bins=None, bbox_coords=None, with_legend=True, title=None, vanilla=False,
plot_mean=False, normed=False, fig_dims=(5, 5)):
"""
Plot a set of samples for Psi of a two isoform gene.
"""
if not fig:
sampled_psi_fig = plt.figure(figsize=fig_dims, dpi=300)
else:
sampled_psi_fig = fig
ax = sampled_psi_fig.add_subplot(num_rows, num_cols, subplot_start)
num_iters = int(self.params['iters'])
burn_in = int(self.params['burn_in'])
lag = int(self.params['lag'])
percent_acceptance = float(self.params['percent_accept'])
proposal_type = self.params['proposal_type']
plt.rcParams['font.size'] = 10
show_spines(ax, ['left', 'bottom'])
bins = bins
assert((value_to_label == None and label == None) or \
(value_to_label != None and label != None))
# retrieve samples
samples_to_plot = self.samples[:, isoform_index]
# picasso blue #0276FD
if not vanilla:
if bins != None:
plt.hist(samples_to_plot, align='mid', lw=0.5, facecolor='#0276FD',
edgecolor='#ffffff')
else:
plt.hist(samples_to_plot, align='mid', lw=0.5, facecolor='#0276FD',
edgecolor='#ffffff')
else:
plt.hist(samples_to_plot, align='mid', facecolor='#0276FD', edgecolor='#0276FD')
plt.xlabel(r'${\hat{\Psi}}_{\mathregular{MISO}}$')
plt.ylabel('Frequency')
plt.xlim([0, 1])
# Normalize samples
if normed:
yticks = list(plt.gca().get_yticks())
print "yticks: ", yticks
ytick_labels = ["%.2f" %(float(ytick) / float(normed)) for ytick in yticks]
ax.set_yticklabels(ytick_labels)
# samples_to_plot = samples_to_plot / float(len(samples_to_plot))
# curr_tick_labels = [label.get_label() for label in ax.get_yticklabels()]
# print "Current tick labels: ", curr_tick_labels
# new_tick_labels = []
# for label in curr_tick_labels:
# if len(label) > 0:
# new_label = "%.1f" %(float(label) / normed)
# else:
# new_label = ""
# new_tick_labels.append(new_label)
# #ax.set_yticklabels(new_tick_labels)
curr_axes = plt.gca()
# Plot MAP estimate for same data
if map_estimate:
l = plt.axvline(x=map_estimate, color='b', linewidth=1.2, ls='-', label=r'${\hat{\Psi}}_{MAP}\ =\ %.2f$' %(map_estimate))
# Plot true Psi
if self.true_psi:
plot_id = "%dsimul_%diters_%dburnin_%dlag_%s_truepsi_%.2f.pdf" \
%(simulation_num, num_iters, burn_in, lag, proposal_type, self.true_psi)
l = plt.axvline(x=self.true_psi, color='r', linewidth=1.2, ls='-', label=r'True $\Psi$')
else:
# Unknown true Psi
plot_id = "%dsimul_%diters_%dburnin_%dlag_%s_%s_truepsi.pdf" \
%(simulation_num, num_iters, burn_in, lag, proposal_type, 'unknown')
if value_to_label:
l = plt.axvline(x=value_to_label, color='r', linewidth=1.2, ls='-', label=label)
# plot credible intervals if given
if plot_intervals:
# print "Plotting %.2f confidence intervals" %(plot_intervals * 100)
interval_c1, interval_c2 = ht.compute_credible_intervals(samples_to_plot, plot_intervals)
plt.axvline(x=interval_c1, color='#999999', linewidth=0.7, ls='--',
label=r'%d' %(plot_intervals*100) + '% CI')
plt.axvline(x=interval_c2, color='#999999', linewidth=0.7, ls='--')
if plot_mean:
sample_mean = mean(samples_to_plot)
plt.axvline(x=sample_mean, color='r', linewidth=0.8, label='Mean')
if with_legend and (plot_intervals or self.true_psi):
if not bbox_coords:
lg = plt.legend(handletextpad=0.172, borderpad=0.01, labelspacing=.008,
handlelength=1.4, loc='best', numpoints=1)
else:
lg = plt.legend(handletextpad=0.172, borderpad=0.01, labelspacing=.008,
handlelength=1.4, loc='best', numpoints=1,
bbox_to_anchor=bbox_coords)
lg.get_frame().set_linewidth(0)
for t in lg.get_texts():
t.set_fontsize(8)
if title:
plt.title(title)
if plots_dir:
if not plot_filename:
plt.savefig(plots_dir + "sampled_psi_hist_%s" %(plot_id))
else:
plt.savefig(plots_dir + plot_filename + '.pdf')
return curr_axes
