def plot(caseHDF5Name, controlHDF5Name, position):
print("Plot theta of position %d" % (position))
###################### plot position for case ###################
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(caseHDF5Name)
casedelta = caseq['delta']
a = casedelta[0][position, 0]
b = casedelta[0][position, 1]
#print (a, b)
fig, ax = plt.subplots()
# display the pdf
# ppf (percentage point function) is the inverse CDF.
# median read depth of case file
cov_case = int(np.median(caseN))
x_case = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax.plot(x_case,
beta.pdf(x_case, a, b),
'b-',
lw=4,
alpha=0.8,
label="Case, Depth=%d" % cov_case)
# generate random variables
r_case = beta.rvs(a, b, size=1000)
ax.hist(r_case, normed=True, histtype='stepfilled', alpha=0.2)
ax.legend(loc='best', frameon=False)
###################### plot position for control ###################
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
controlHDF5Name)
controldelta = controlq['delta']
a = controldelta[0][position, 0]
b = controldelta[0][position, 1]
#print (a, b)
# display the pdf
# ppf (percentage point function) is the inverse CDF.
cov_control = int(np.median(controlN))
x_control = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax.plot(x_control,
beta.pdf(x_control, a, b),
'g-',
lw=4,
alpha=0.8,
label='Control, Depth=%d' % cov_control)
# generate random variables
r_control = beta.rvs(a, b, size=1000)
ax.hist(r_control, normed=True, histtype='stepfilled', alpha=0.2)
ax.legend(loc='best', frameon=False)
ax.set_title(
'$\\beta$ variational distribution of $\\theta$ at position %d' %
(position))
ax.set_xlabel('$\\theta$', fontsize=20)
ax.set_ylabel('PDF', fontsize=18)
# position_VAF_downsample
#plt.savefig('%d.png' %(position))
plt.show()
def main():
dilutionList = (0.1, 0.3, 1.0, 10.0, 100.0)
folder = '2015-09-28_Run_rvd3_synthetic_data_set/hdf5/10'
fig = plt.figure(figsize=(12, 20))
#plt.suptitle('Read depth/M across position')
controlFile = "../%s/Control.hdf5" % folder
controlR, controlN, controlPhi, controlq, controlLoc, _ = rvd3.load_model(
controlFile)
sub0 = len(dilutionList) + 1
ax = fig.add_subplot(sub0, 2, 1)
#TODO: use index of controlN rather than directly controlLOC
controlLoc = [int(x.split(':')[1]) for x in controlLoc]
ax.plot(controlLoc, controlN.T)
ax.set_title('Control')
ax.set_ylabel('Coverage')
ax.ticklabel_format(style='sci', axis='y', scilimits=(0, 0))
ax = fig.add_subplot(sub0, 2, 2)
ax.semilogy(controlLoc, np.mean(controlPhi['M'], axis=1))
ax.set_title('Control')
ax.set_ylabel('M')
#ax.set_ylim([1e-4,1e5])
ax.semilogy([controlLoc[0], controlLoc[-1]],
[controlPhi['M0'], controlPhi['M0']],
color='r',
ls='--')
for d in dilutionList:
logging.debug("Processing dilution: %0.1f%%" % d)
caseFile = "Case%s.hdf5" % str(d).replace(".", "_")
caseFile = "../%(folder)s/%(file)s" % {
'folder': folder,
'file': caseFile
}
caseR, caseN, casePhi, caseq, caseLoc, _ = rvd3.load_model(caseFile)
ax = fig.add_subplot(sub0, 2, 2 * dilutionList.index(d) + 3)
caseLoc = [int(x.split(':')[1]) for x in caseLoc]
ax.plot(caseLoc, caseN.T)
ax.set_title("Dilution %0.1f%%" % d)
if dilutionList.index(d) == len(dilutionList) - 1:
ax.set_xlabel('Position')
ax.set_ylabel('Coverage')
ax.ticklabel_format(style='sci', axis='y', scilimits=(0, 0))
ax = fig.add_subplot(sub0, 2, 2 * dilutionList.index(d) + 4)
ax.semilogy(caseLoc, np.mean(casePhi['M'], axis=1))
ax.set_title("Dilution %0.1f%%" % d)
if dilutionList.index(d) == len(dilutionList) - 1:
ax.set_xlabel('Position')
ax.set_ylabel('M')
#ax.set_ylim([1e-4,1e5])
ax.semilogy([caseLoc[0], caseLoc[-1]], [casePhi['M0'], casePhi['M0']],
color='r',
ls='--')
plt.savefig('M_dsample=10.png')
def bayestest(caseHDF5Name, controlHDF5Name, position):
alpha = 0.05
tau = 0
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(caseHDF5Name)
casegam = caseq['gam']
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
controlHDF5Name)
controlgam = controlq['gam']
def beta_mean(p):
return p[0] * 1.0 / np.sum(p)
def beta_var(p):
s = np.sum(p)
return p[0] * p[1] / (s**2 * (s + 1))
mu = (beta_mean(casegam[position, :]) - casephi['mu0']) - (
beta_mean(controlgam[position, :]) - controlphi['mu0'])
sigma = beta_var(casegam[position, :]) + beta_var(controlgam[position, :])
z = (tau - mu) / sigma
print(z)
p = ss.norm.cdf(z)
print(p[0])
def read(filename, pos):
def beta_mean(p):
return p[0] * 1.0 / np.sum(p)
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(filename)
casegam = caseq['gam']
caseMu = beta_mean(casegam[pos, :]) - casephi['mu0']
# calculate the lower and upper credible value
alpha = 0.05
cred = caseMu * alpha / 2
conf_l = caseMu - cred
conf_u = caseMu + cred
# calculate the error bar value
err = np.array(cred, cred)
print 100 * caseMu, conf_l, conf_u
return 100 * caseMu, 100 * err
def main():
book=xlwt.Workbook(encoding="utf-8")
sheet1=book.add_sheet("TPR_TNR")
sheet1.write(0, 0, "VAF")
sheet1.write(0, 1, "Median Depth")
sheet2=book.add_sheet("Multi-measures")
sheet2.write(1, 0, "VAF")
sheet2.write(1, 1, "Median Depth")
sheet3=book.add_sheet("FDR")
sheet3.write(0, 0, "VAF")
sheet3.write(0, 1, "Median Depth")
sheet4=book.add_sheet("MCC")
sheet4.write(0, 0, "VAF")
sheet4.write(0, 1, "Median Depth")
# method = {'RVD2(T*)(R=6)':'./../2013-12-20_experiment_set_gibbs_Qsd_mu_1_mu_over_10_minus_mu0/six_replicates_synthetic_optT/vcf/MCC',
# 'RVD2(T*)(R=1)':'./../2013-12-20_experiment_set_gibbs_Qsd_mu_1_mu_over_10_minus_mu0/one_replicate_synthetic_optT/vcf/MCC',
# 'RVD2(T=0)(R=6)':'./../2013-12-20_experiment_set_gibbs_Qsd_mu_1_mu_over_10_minus_mu0/six_replicates_synthetic_T0/vcf',
# 'RVD2(T=0)(R=1)':'./../2013-12-20_experiment_set_gibbs_Qsd_mu_1_mu_over_10_minus_mu0/one_replicate_synthetic_T0/vcf',
# 'VarScan2 somatic':'./../2013-09-23_SNP_calling_using_varscan2_somatic/vcf',
# 'SAMtools':'./../2013-09-10_SNP_calling_using_samtools/vcf',
# 'GATK':'./../2013-09-13_SNP_calling_using_GATK/vcf',
# 'MuTect':'./../2013-10-02_SNP_calling_using_MuTect/work',
# 'Strelka':'./../2013-10-01_SNP_calling_using_strelka/vcf',
# 'VarScan2 mpileup':'./../2013-09-20_SNP_calling_using_varscan2/vcf'}
# method = {'RVD2_MCMC(T=0,R=6)':'./../2013-12-20_experiment_set_gibbs_Qsd_mu_1_mu_over_10_minus_mu0/six_replicates_synthetic_T0/vcf',
# 'MuTect':'./../2013-10-02_SNP_calling_using_MuTect/work',
# 'RVD2_Var(T=0,R=6)':'./vcf'
# }
method = {'RVD3(T=0,R=6)':'./vcf',}
DilutionList = (0.1, 0.3, 1.0, 10.0,100.0)
DepthList = (10000, 1000, 100, 10)
i=0
for k, v in method.iteritems():
i=i+1
print 'Method %(number)d: %(method)s' %{'number':i, 'method': k}
sheet1.write(0, i+1, k)
sheet2.write(0, 9*(i-1)+6, k)
sheet3.write(0, i+1, k)
sheet4.write(0, i+1, k)
character=('Sensitiviy', 'Specificity', 'FPR', 'FNR', 'PPV', 'NPV', 'FDR', 'ACC', 'MCC')
for j in xrange(9):
sheet2.write(1,9*(i-1)+j+2,character[j])
for d in DilutionList:
if i==1:
sheet1.write(DilutionList.index(d)*len(DepthList)+1,0,"%0.1f%%" %d)
sheet2.write(DilutionList.index(d)*len(DepthList)+2,0,"%0.1f%%" %d)
sheet3.write(DilutionList.index(d)*len(DepthList)+1,0,"%0.1f%%" %d)
sheet4.write(DilutionList.index(d)*len(DepthList)+1,0,"%0.1f%%" %d)
for r in DepthList:
# read in the median coverage
#hdf5Dir='../2013-08-14_Compute_ROC_Synthetic_avg%s' %str(r)
hdf5Dir = './hdf5/%s' %str(r)
caseFile = 'Case%s.hdf5' %str(d).replace('.','_')
caseFile = "%(dir)s/%(file)s" %{'dir':hdf5Dir,'file':caseFile}
# pdb.set_trace()
#(_, _, _, _, _, caseN,_) = rvd27.load_model(caseFile)
(_, caseN, _, _, _, _) = rvd3.load_model(caseFile)
cov = int(np.median(caseN))
# print the median coverage
if i==1:
sheet1.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+1, 1, "%s" % str(cov))
sheet2.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+2, 1, "%s" % str(cov))
sheet3.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+1, 1, "%s" % str(cov))
sheet4.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+1, 1, "%s" % str(cov))
# read in called positions from vcf files
# pdb.set_trace()
vcfFile=os.path.join(v,"%s" %r,
"vcf%s.vcf" %str(d).replace('.','_'))
logging.debug(vcfFile)
vcf_reader = vcf.Reader(open(vcfFile, 'r'))
# pdb.set_trace()
callpos=np.array([record.POS for record in vcf_reader])
# prediction classification
PredictClass = np.zeros(400)
if len(callpos) != 0:
PredictClass[callpos-1] = np.ones_like(callpos)
# actual classification
RefClass = np.zeros(400)
pos = np.arange(85,346,20)
RefClass[pos-1] = np.ones_like(pos)
# characteristics computation
[TPR, TNR, FPR, FNR, PPV, NPV, FDR, ACC, MCC]=characteristics(RefClass, PredictClass)
ncharacter=(TPR, TNR, FPR, FNR, PPV, NPV, FDR, ACC, MCC)
# print characteristics
sheet1.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+1,i+1,"%(TPR)0.2f/%(TNR)0.2f" %{'TPR':TPR,'TNR':TNR})
for j in xrange(9):
sheet2.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+2,9*(i-1)+j+2,'%0.2f' %ncharacter[j])
if not np.isnan(FDR):
sheet3.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+1,i+1,"%0.2f" %FDR)
if not np.isnan(FDR):
sheet4.write(DilutionList.index(d)*len(DepthList)+DepthList.index(r)+1,i+1,"%0.2f" %MCC)
book.save('statistics_no_chi2.xls')
def plot(case_10000, case_1000, case_100, case_10, control_10000, control_1000,
control_100, control_10, position):
print("Plot mu of position %d" % (position + 1))
fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(nrows=2, ncols=2)
xstep = 0.002
xmax = 0.0152
ymax = 1200
size = 15
################ Downsample = 10000 #################################################
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(case_10000)
casegam = caseq['gam']
a, b = get_a_b(position, casegam, casephi['mu0'])
cov_case = int(np.median(caseN))
x_case = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax1.plot(x_case,
beta.pdf(x_case, a, b),
'b-',
lw=5,
alpha=0.8,
label="Case")
# generate random variables
r_case = beta.rvs(a, b, size=1000)
ax1.hist(r_case, normed=True, histtype='stepfilled', alpha=0.2)
ax1.legend(loc='best', frameon=False)
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
control_10000)
controlgam = controlq['gam']
a, b = get_a_b(position, controlgam, controlphi['mu0'])
x_control = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax1.plot(x_control,
beta.pdf(x_control, a, b),
'g-',
lw=5,
alpha=0.8,
label='Control')
# generate random variables
r_control = beta.rvs(a, b, size=1000)
ax1.hist(r_control, normed=True, histtype='stepfilled', alpha=0.2)
ax1.legend(loc='best', frameon=False)
ax1.set_title('Depth=%d' % cov_case, fontsize=size)
xticks = np.arange(0, xmax, xstep)
ax1.set_xticks(xticks)
ax1.set_ylim(0, ymax)
print('mu0^control:', controlphi['mu0'])
print('mu0^case', casephi['mu0'], '\n')
#################### Downsample = 1000 #############################################
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(case_1000)
casegam = caseq['gam']
a, b = get_a_b(position, casegam, casephi['mu0'])
cov_case = int(np.median(caseN))
x_case = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax2.plot(x_case,
beta.pdf(x_case, a, b),
'b-',
lw=5,
alpha=0.8,
label="Case")
# generate random variables
r_case = beta.rvs(a, b, size=1000)
ax2.hist(r_case, normed=True, histtype='stepfilled', alpha=0.2)
ax2.legend(loc='best', frameon=False)
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
control_1000)
controlgam = controlq['gam']
a, b = get_a_b(position, controlgam, controlphi['mu0'])
x_control = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax2.plot(x_control,
beta.pdf(x_control, a, b),
'g-',
lw=5,
alpha=0.8,
label='Control')
# generate random variables
r_control = beta.rvs(a, b, size=1000)
ax2.hist(r_control, normed=True, histtype='stepfilled', alpha=0.2)
ax2.legend(loc='best', frameon=False)
ax2.set_title('Depth=%d' % cov_case, fontsize=size)
xticks = np.arange(0, xmax, xstep)
ax2.set_xticks(xticks)
ax2.set_ylim(0, ymax)
print('mu0^control:', controlphi['mu0'])
print('mu0^case', casephi['mu0'], '\n')
################### Downsample = 100 #############################################
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(case_100)
casegam = caseq['gam']
a, b = get_a_b(position, casegam, casephi['mu0'])
cov_case = int(np.median(caseN))
x_case = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax3.plot(x_case,
beta.pdf(x_case, a, b),
'b-',
lw=5,
alpha=0.8,
label="Case")
# generate random variables
r_case = beta.rvs(a, b, size=1000)
ax3.hist(r_case, normed=True, histtype='stepfilled', alpha=0.2)
ax3.legend(loc='best', frameon=False)
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
control_100)
controlgam = controlq['gam']
a, b = get_a_b(position, controlgam, controlphi['mu0'])
x_control = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax3.plot(x_control,
beta.pdf(x_control, a, b),
'g-',
lw=5,
alpha=0.8,
label='Control')
# generate random variables
r_control = beta.rvs(a, b, size=1000)
ax3.hist(r_control, normed=True, histtype='stepfilled', alpha=0.2)
ax3.legend(loc='best', frameon=False)
ax3.set_title('Depth=%d' % cov_case, fontsize=size)
xticks = np.arange(0, xmax, xstep)
ax3.set_xticks(xticks)
ax3.set_xlabel('$\mu$', fontsize=size)
ax3.set_ylim(0, ymax)
print('mu0^control:', controlphi['mu0'])
print('mu0^case', casephi['mu0'], '\n')
#################### Downsample = 10 #############################################
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(case_10)
casegam = caseq['gam']
a, b = get_a_b(position, casegam, casephi['mu0'])
cov_case = int(np.median(caseN))
x_case = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax4.plot(x_case,
beta.pdf(x_case, a, b),
'b-',
lw=5,
alpha=0.8,
label="Case")
# generate random variables
r_case = beta.rvs(a, b, size=1000)
ax4.hist(r_case, normed=True, histtype='stepfilled', alpha=0.2)
ax4.legend(loc='best', frameon=False)
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
control_10)
controlgam = controlq['gam']
a, b = get_a_b(position, controlgam, controlphi['mu0'])
x_control = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax4.plot(x_control,
beta.pdf(x_control, a, b),
'g-',
lw=5,
alpha=0.8,
label='Control')
# generate random variables
r_control = beta.rvs(a, b, size=1000)
ax4.hist(r_control, normed=True, histtype='stepfilled', alpha=0.2)
ax4.legend(loc='best', frameon=False)
ax4.set_title('Depth=%d' % cov_case, fontsize=size)
xticks = np.arange(0, xmax, xstep)
ax4.set_xticks(xticks)
ax4.set_xlabel('$\mu$', fontsize=size)
ax4.set_ylim(0, ymax)
print('mu0^control:', controlphi['mu0'])
print('mu0^case', casephi['mu0'], '\n')
plt.suptitle(
'$\\beta$ variational distribution of $\mu$ at position %d when VAF=1.0%% '
% (position + 1),
fontsize=size)
# manually adjust the spacing of suptitle
plt.subplots_adjust(top=0.9)
#plt.tight_layout(fig, rect=[0, 0.03, 1, 0.95])
#plt.show()
fig = plt.gcf()
fig.set_size_inches(12, 8)
plt.savefig('mu_%d_VAF=1.0.png' % (position + 1))
def plot(f2HDF5Name, f1HDF5Name, f3HDF5Name, f4HDF5Name, position):
###################### plot position for f2 ###################
f2R, f2N, f2phi, f2q, loc, refb = rvd3.load_model(f2HDF5Name)
f2gam = f2q['gam']
a = f2gam[position, 0]
b = f2gam[position, 1]
#print (a, b)
M0_f2 = f2phi['M0']
fig, ax = plt.subplots()
x_f2 = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax.plot(x_f2,
beta.pdf(x_f2, a, b),
'g--',
lw=8,
alpha=0.8,
label="M0=%.4f" % M0_f2)
###################### plot position for f1 ###################
f1R, f1N, f1phi, f1q, _, _ = rvd3.load_model(f1HDF5Name)
f1gam = f1q['gam']
a = f1gam[position, 0]
b = f1gam[position, 1]
M0_f1 = f1phi['M0']
x_f1 = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax.plot(x_f1,
beta.pdf(x_f1, a, b),
'b-',
lw=8,
alpha=0.8,
label='M0=%.3f' % M0_f1)
###################### plot position for f3 ###################
f3R, f3N, f3phi, f3q, _, _ = rvd3.load_model(f3HDF5Name)
f3gam = f3q['gam']
a = f3gam[position, 0]
b = f3gam[position, 1]
M0_f3 = f3phi['M0']
x_f3 = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax.plot(x_f3,
beta.pdf(x_f3, a, b),
'm-.',
lw=8,
alpha=0.8,
label='M0=%.2f' % M0_f3)
###################### plot position for f4 ###################
f4R, f4N, f4phi, f4q, _, _ = rvd3.load_model(f4HDF5Name)
f4gam = f4q['gam']
a = f4gam[position, 0]
b = f4gam[position, 1]
M0_f4 = f4phi['M0']
x_f4 = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
ax.plot(x_f4,
beta.pdf(x_f4, a, b),
'yo:',
lw=8,
alpha=0.8,
label='M0=%.1f' % M0_f4)
legend = ax.legend(loc='upper left')
for label in legend.get_texts():
label.set_fontsize(38)
ax.set_xlabel('$\hat{\mu}_{1,014,740}$', fontsize=38)
ax.set_xlim([0.95, 1])
plt.setp(plt.gca().get_xticklabels(), fontsize=35)
plt.setp(plt.gca().get_yticklabels(), fontsize=35)
plt.show()
def main():
################### Read mu of MCMC (rvd2) ################################
with h5py.File(control_mcmc, 'r') as f:
muControl = f['mu'][...]
locControl = f['loc'][...]
with h5py.File(case_mcmc, 'r') as f:
muCase = f['mu'][...]
locCase = f['loc'][...]
idx = []
for pos in position:
idx.append(pos)
muControl1 = muControl[idx]
muCase1 = muCase[idx]
#N = 2000
#(muZ,_,_) =rvd27.sample_post_diff(muCase1, muControl1, N) # sample Z
## plot histogram
num_bins = 25
for i in xrange(len(position)):
fig = plt.figure(figsize=(12, 8))
########### Plot mu of MCMC (rvd2) vs Variational (rvd3) ##################
# normed=True, the integral of the histogram will sum to 1.
plt.hist(muCase1[i, :].T,
num_bins,
normed=True,
facecolor='r',
alpha=0.5,
label='Case (MCMC)')
plt.hist(muControl1[i, :].T,
num_bins,
normed=True,
facecolor='k',
alpha=0.5,
label='Control (MCMC)')
############# Plot mu of Variational (rvd3) ################################
caseR, caseN, casephi, caseq, loc, refb = rvd3.load_model(case_var)
casegam = caseq['gam']
a = casegam[position, 0]
b = casegam[position, 1]
cov_case = int(np.median(caseN))
x_case = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b), 100)
plt.plot(x_case,
beta.pdf(x_case, a, b),
'r--',
lw=4,
alpha=1.0,
label="Case (Variational)")
r_case = beta.rvs(a, b, size=2000)
plt.hist(r_case,
num_bins,
normed=True,
histtype='stepfilled',
alpha=0.2,
facecolor='r')
controlR, controlN, controlphi, controlq, _, _ = rvd3.load_model(
control_var)
controlgam = controlq['gam']
a = controlgam[position, 0]
b = controlgam[position, 1]
x_control = np.linspace(beta.ppf(0.001, a, b), beta.ppf(0.999, a, b),
100)
plt.plot(x_control,
beta.pdf(x_control, a, b),
'k--',
lw=4,
alpha=1.0,
label='Control (Variational)')
r_control = beta.rvs(a, b, size=2000)
plt.hist(r_control,
num_bins,
normed=True,
histtype='stepfilled',
alpha=0.2,
facecolor='k')
plt.xlim(0, 0.012)
plt.legend(loc='best', frameon=False)
plt.xlabel('$\hat{\mu} = \mu-\mu_0$', fontsize=20)
plt.xticks(rotation=25)
plt.title('$\hat{\mu}$ at position %s when median depth is %d' %
((position[i] + 1), cov_case),
fontsize=18)
plt.xticks(rotation=25)
plt.savefig('position_%s_%d_mcmc_vs_var.png' %
((position[i] + 1), cov_case))
plt.tight_layout()
def ROCpoints(controlFile,caseFile, d, N, P, chi2):
# Load the model samples
controlR, controlN, controlphi, controlq, controlLoc, _ = rvd3.load_model(controlFile)
controlgam = controlq['gam']
caseR, caseN, casephi, caseq, caseLoc, refb = rvd3.load_model(caseFile)
casegam = caseq['gam']
#(N,J) = np.shape(caseR)[0:2]
J = len(controlLoc)
def beta_mean(p):
return p[0]*1.0/np.sum(p)
def beta_var(p):
s = np.sum(p)
return p[0]*p[1]/(s**2*(s+1))
# Draw random samples from Beta distribution
controlMu = np.zeros(shape=(J, 4000))
caseMu = np.zeros(shape=(J, 4000))
for j in xrange(J):
controlMu[j] = np.random.beta(controlgam[j,:][0], controlgam[j,:][1], 4000)
caseMu[j] = np.random.beta(casegam[j,:][0], casegam[j,:][1], 4000)
# Extract the common locations in case and control
caseLocIdx = [i for i in xrange(len(caseLoc)) if caseLoc[i] in controlLoc]
controlLocIdx = [i for i in xrange(len(controlLoc)) if controlLoc[i] in caseLoc]
caseMu = caseMu[caseLocIdx,:]
controlMu = controlMu[controlLocIdx,:]
# caseR = caseR[:,caseLocIdx,:]
# controlR = controlR[:,controlLocIdx,:]
# caseN = caseN[:,caseLocIdx]
# controlN = controlN[:,controlLocIdx]
loc = caseLoc[caseLocIdx]
J = len(loc)
pos = np.arange(85,346,20)
posidx = [i for i in xrange(J) if int(loc[i][8:]) in pos]
# Sample from the posterior Z = muCase - muControl
(Z, caseMuS, controlMuS) = sample_post_diff(caseMu-casephi['mu0'], controlMu-controlphi['mu0'], N)
# Compute cumulative posterior probability for regions (Threshold,np.inf)
T = np.linspace(np.min(np.min(Z)), np.max(np.max(Z)), num=300)
pList = [bayes_test(Z, [(t, np.inf)]) for t in T]
# mutation classification
clsList = np.array((np.array(pList)>P).astype(int))
clsList = clsList.reshape((clsList.shape[0],clsList.shape[1]))# category list
# chi2 test for goodness-of-fit to a uniform distribution for non-ref bases
if chi2:
nRep = caseR.shape[0]
chi2Prep = np.zeros((J,nRep))
chi2P = np.zeros(J)
for j in xrange(J):
chi2Prep[j,:] = np.array([rvd3.chi2test( caseR[i,j,:] ) for i in xrange(nRep)] )
if np.any(np.isnan(chi2Prep[j,:])):
chi2P[j] = 1
else:
chi2P[j] = 1-ss.chi2.cdf(-2*np.sum(np.log(chi2Prep[j,:] + np.finfo(float).eps)), 2*nRep) # combine p-values using Fisher's Method
clsList2 = np.array((np.array(chi2P)<0.05/J).astype(int))
clsList2 = np.tile(clsList2,(clsList.shape[0],1))
clsList = np.array(((clsList+clsList2)==2).astype(int))
# false postive rate
fpr = np.array([float(sum(clsList[i])-sum(clsList[i,np.array(posidx)]))/(clsList.shape[1]-len(posidx)) for i in xrange(clsList.shape[0])])
# true positive rate
tpr = np.array([float(sum(clsList[i,np.array(posidx)]))/len(posidx) for i in xrange(clsList.shape[0])])
cov = np.median(caseN)
# # return information for mu bar plot at called positions under optimal threshold.
# # using EL distance.
# ## distance=np.sum(np.power([fpr,tpr-1],2),0)
# ## Tidx=distance.argmin()
# ## print Tidx
# # Using L1 distance
# distance = 1+tpr-fpr
# Tidx=distance.argmax()
# outputFile=os.path.join(path,'vcf%s.vcf' %str(d).replace('.','_'))
# with h5py.File(controlFile, 'r') as f:
# refb = f['/refb'][...]
# f.close()
# refb = refb[controlLocIdx]
# altb = []
# call=[]
# acgt = {'A':0, 'C':1, 'G':2, 'T':3}
# for i in xrange(J):
# r = np.squeeze(caseR[:,i,:]) # replicates x bases
# # Make a list of the alternate bases for each replicate
# acgt_r = ['A','C','G','T']
# del acgt_r[ acgt[refb[i]] ]
# altb_r = [acgt_r[x] for x in np.argmax(r, axis=1)]
# if clsList[Tidx,i]==1:
# call.append(True)
# altb.append(altb_r[0])
# else:
# altb.append(None)
# call.append(False)
# rvd30.write_vcf(outputFile, loc, call, refb, altb, np.mean(caseMu, axis=1), np.mean(controlMu, axis=1))
return fpr,tpr, cov
