def Run(self):
self.transit_message("Starting Genetic Interactions Method")
start_time = time.time()
self.output.write("#GI\n")
wiglist = self.ctrldataA + self.expdataA + self.ctrldataB + self.expdataB
Nwig = len(wiglist)
Na1 = len(self.ctrldataA)
Nb1 = len(self.expdataA)
Na2 = len(self.ctrldataB)
Nb2 = len(self.expdataB)
# Get data
self.transit_message("Getting Data")
(data, position) = transit_tools.get_validated_data(wiglist, wxobj=self.wxobj)
# Normalize data if specified
if self.normalization != "nonorm":
self.transit_message("Normalizing using: %s" % self.normalization)
(data, factors) = norm_tools.normalize_data(data, self.normalization, wiglist, self.annotation_path)
# Do LOESS correction if specified
if self.LOESS:
self.transit_message("Performing LOESS Correction")
for j in range(K):
data[j] = stat_tools.loess_correction(position, data[j])
# Get Gene objects for each condition
G_A1 = tnseq_tools.Genes([], self.annotation_path, data=data[:Na1], position=position,nterm=self.NTerminus,cterm=self.CTerminus)
G_B1 = tnseq_tools.Genes([], self.annotation_path, data=data[Na1:(Na1+Nb1)], position=position,nterm=self.NTerminus,cterm=self.CTerminus)
G_A2 = tnseq_tools.Genes([], self.annotation_path, data=data[(Na1+Nb1):(Na1+Nb1+Na2)], position=position,nterm=self.NTerminus,cterm=self.CTerminus)
G_B2 = tnseq_tools.Genes([], self.annotation_path, data=data[(Na1+Nb1+Na2):], position=position,nterm=self.NTerminus,cterm=self.CTerminus)
means_list_a1 = []
means_list_b1 = []
means_list_a2 = []
means_list_b2 = []
var_list_a1 = []
var_list_a2 = []
var_list_b1 = []
var_list_b2 = []
# Base priors on empirical observations accross genes.
for gene in sorted(G_A1):
if gene.n > 1:
A1_data = G_A1[gene.orf].reads.flatten()
B1_data = G_B1[gene.orf].reads.flatten()
A2_data = G_A2[gene.orf].reads.flatten()
B2_data = G_B2[gene.orf].reads.flatten()
means_list_a1.append(numpy.mean(A1_data))
var_list_a1.append(numpy.var(A1_data))
means_list_b1.append(numpy.mean(B1_data))
var_list_b1.append(numpy.var(B1_data))
means_list_a2.append(numpy.mean(A2_data))
var_list_a2.append(numpy.var(A2_data))
means_list_b2.append(numpy.mean(B2_data))
var_list_b2.append(numpy.var(B2_data))
# Priors
mu0_A1 = scipy.stats.trim_mean(means_list_a1, 0.01)
mu0_B1 = scipy.stats.trim_mean(means_list_b1, 0.01)
mu0_A2 = scipy.stats.trim_mean(means_list_a2, 0.01)
mu0_B2 = scipy.stats.trim_mean(means_list_b2, 0.01)
s20_A1 = scipy.stats.trim_mean(var_list_a1, 0.01)
s20_B1 = scipy.stats.trim_mean(var_list_b1, 0.01)
s20_A2 = scipy.stats.trim_mean(var_list_a2, 0.01)
s20_B2 = scipy.stats.trim_mean(var_list_b2, 0.01)
k0=1.0
nu0=1.0
data = []
postprob = []
count = 0
N = len(G_A1)
self.progress_range(N)
# Perform actual analysis
for gene in G_A1:
# If there is some data
if gene.n > 0:
A1_data = G_A1[gene.orf].reads.flatten()
B1_data = G_B1[gene.orf].reads.flatten()
A2_data = G_A2[gene.orf].reads.flatten()
B2_data = G_B2[gene.orf].reads.flatten()
# Time-1 Time-2
#
# Strain-A A C
#
# Strain-B B D
try:
muA1_post, varA1_post = stat_tools.sample_trunc_norm_post(A1_data, self.samples,
mu0_A1, s20_A1, k0, nu0)
muB1_post, varB1_post = stat_tools.sample_trunc_norm_post(B1_data, self.samples,
mu0_B1, s20_B1, k0, nu0)
muA2_post, varA2_post = stat_tools.sample_trunc_norm_post(A2_data, self.samples,
mu0_A2, s20_A2, k0, nu0)
muB2_post, varB2_post = stat_tools.sample_trunc_norm_post(B2_data, self.samples,
mu0_B2, s20_B2, k0, nu0)
except Exception as e:
muA1_post = varA1_post = numpy.ones(self.samples)
muB1_post = varB1_post = numpy.ones(self.samples)
muA2_post = varA2_post = numpy.ones(self.samples)
muB2_post = varB2_post = numpy.ones(self.samples)
logFC_A_post = numpy.log2(muA2_post/muA1_post)
logFC_B_post = numpy.log2(muB2_post/muB1_post)
delta_logFC_post = logFC_B_post - logFC_A_post
alpha = 0.05
# Get Bounds of the HDI
l_logFC_A, u_logFC_A = stat_tools.HDI_from_MCMC(logFC_A_post, 1-alpha)
l_logFC_B, u_logFC_B = stat_tools.HDI_from_MCMC(logFC_B_post, 1-alpha)
l_delta_logFC, u_delta_logFC = stat_tools.HDI_from_MCMC(delta_logFC_post, 1-alpha)
mean_logFC_A = numpy.mean(logFC_A_post)
mean_logFC_B = numpy.mean(logFC_B_post)
mean_delta_logFC = numpy.mean(delta_logFC_post)
# Is HDI significantly different than ROPE?
not_HDI_overlap_bit = l_delta_logFC > self.rope or u_delta_logFC < -self.rope
# Probability of posterior overlaping with ROPE
probROPE = numpy.mean(numpy.logical_and(delta_logFC_post>=0.0-self.rope, delta_logFC_post<=0.0+self.rope))
# If there is no data, assume empty defaults
else:
A1_data = [0,0]
B1_data = [0,0]
A2_data = [0,0]
B2_data = [0,0]
muA1_post = varA1_post = numpy.ones(self.samples)
muB1_post = varB1_post = numpy.ones(self.samples)
muA2_post = varA2_post = numpy.ones(self.samples)
muB2_post = varB2_post = numpy.ones(self.samples)
logFC_A_post = numpy.log2(muA2_post/muA1_post)
logFC_B_post = numpy.log2(muB2_post/muB1_post)
delta_logFC_post = logFC_B_post - logFC_A_post
mean_logFC_A = 0
mean_logFC_B = 0
mean_delta_logFC = 0
l_logFC_A = 0
u_logFC_A = 0
l_logFC_B = 0
u_logFC_B = 0
l_delta_logFC = 0
u_delta_logFC = 0
probROPE = 1.0
if numpy.isnan(l_logFC_A):
l_logFC_A = -10
u_logFC_A = 10
if numpy.isnan(l_logFC_B):
l_logFC_B = -10
u_logFC_B = 10
if numpy.isnan(l_delta_logFC):
l_delta_logFC = -10
u_delta_logFC = 10
postprob.append(probROPE)
data.append((gene.orf, gene.name, gene.n, numpy.mean(muA1_post), numpy.mean(muA2_post), numpy.mean(muB1_post), numpy.mean(muB2_post), mean_logFC_A, mean_logFC_B, mean_delta_logFC, l_delta_logFC, u_delta_logFC, probROPE, not_HDI_overlap_bit))
text = "Running GI Method... %2.0f%%" % (100.0*(count+1)/N)
self.progress_update(text, count)
self.transit_message_inplace("Running Export Method... %1.1f%%" % (100.0*count/(N-1)))
count+=1
data.sort(key=lambda x: x[-2])
if self.doBFDR or not self.doFWER:
postprob = numpy.array(postprob)
postprob.sort()
bfdr = numpy.cumsum(postprob)/numpy.arange(1, len(postprob)+1)
adjusted_prob = bfdr
adjusted_label = "BFDR"
elif doFWER:
fwer = FWER_Bayes(postprob)
fwer.sort()
adjusted_prob = fwer
adjusted_label = "FWER"
# If not using adjustment for classification, sort correctly
if not self.doBFDR and not self.doFWER:
sorted_index = numpy.argsort([d[-1] for d in data])[::-1][:len(data)]
adjusted_prob = [adjusted_prob[ii] for ii in sorted_index]
data = [data[ii] for ii in sorted_index]
# Print output
if self.wxobj:
members = sorted([attr for attr in dir(self) if not callable(getattr(self,attr)) and not attr.startswith("__")])
memberstr = ""
for m in members:
memberstr += "%s = %s, " % (m, getattr(self, m))
self.output.write("#GUI with: norm=%s, samples=%s, includeZeros=%s, output=%s\n" % (self.normalization, self.samples, self.includeZeros, self.output.name.encode('utf-8')))
else:
self.output.write("#Console: python %s\n" % " ".join(sys.argv))
self.output.write("#Control Data-A: %s\n" % (",".join(self.ctrldataA).encode('utf-8')))
self.output.write("#Control Data-B: %s\n" % (",".join(self.ctrldataB).encode('utf-8')))
self.output.write("#Experimental Data-A: %s\n" % (",".join(self.expdataA).encode('utf-8')))
self.output.write("#Experimental Data-B: %s\n" % (",".join(self.expdataB).encode('utf-8')))
self.output.write("#Annotation path: %s\n" % (self.annotation_path.encode('utf-8')))
self.output.write("#Time: %s\n" % (time.time() - start_time))
self.output.write("#%s\n" % "\t".join(columns))
if self.doBFDR or self.doFWER:
self.output.write("# Significant interactions are those whose adjusted probability of the delta-logFC falling within ROPE is < 0.05 (Adjusted using %s)\n" % (adjusted_label))
else:
self.output.write("# Significant interactions are those genes whose delta-logFC HDI does not overlap the ROPE\n")
self.output.write("#\n")
# Write column names
self.output.write("#ORF\tName\tNumber of TA Sites\tMean count (Strain A Time 1)\tMean count (Strain A Time 2)\tMean count (Strain B Time 1)\tMean count (Strain B Time 2)\tMean logFC (Strain A)\tMean logFC (Strain B) \tMean delta logFC\tLower Bound delta logFC\tUpper Bound delta logFC\tProb. of delta-logFC being within ROPE\tAdjusted Probability (%s)\tIs HDI outside ROPE?\tType of Interaction\n" % adjusted_label)
# Write gene results
for i,row in enumerate(data):
#1 2 3 4 5 6 7 8 9 10 11 12 13 14
orf, name, n, mean_muA1_post, mean_muA2_post, mean_muB1_post, mean_muB2_post, mean_logFC_A, mean_logFC_B, mean_delta_logFC, l_delta_logFC, u_delta_logFC, probROPE, not_HDI_overlap_bit = row
type_of_interaction = "No Interaction"
if ((self.doBFDR or self.doFWER) and adjusted_prob[i] < 0.05):
type_of_interaction = self.classify_interaction(mean_delta_logFC, mean_logFC_B, mean_logFC_A)
elif not (self.doBFDR or self.doFWER) and not_HDI_overlap_bit:
type_of_interaction = self.classify_interaction(mean_delta_logFC, mean_logFC_B, mean_logFC_A)
new_row = tuple(list(row[:-1])+[adjusted_prob[i], not_HDI_overlap_bit, type_of_interaction])
self.output.write("%s\t%s\t%d\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.8f\t%1.8f\t%s\t%s\n" % new_row)
self.transit_message("Adding File: %s" % (self.output.name))
self.add_file(filetype="GI")
self.finish()
self.transit_message("Finished Genetic Interactions Method")
def Run(self):
self.transit_message("Starting Genetic Interactions Method")
start_time = time.time()
self.output.write("#GI\n")
wiglist = self.ctrldataA + self.ctrldataB + self.expdataA + self.expdataB
Nwig = len(wiglist)
Na1 = len(self.ctrldataA)
Nb1 = len(self.ctrldataB)
Na2 = len(self.expdataA)
Nb2 = len(self.expdataB)
# Get data
self.transit_message("Getting Data")
(data, position) = transit_tools.get_validated_data(wiglist,
wxobj=self.wxobj)
# Normalize data if specified
if self.normalization != "nonorm":
self.transit_message("Normalizing using: %s" % self.normalization)
(data,
factors) = norm_tools.normalize_data(data, self.normalization,
wiglist,
self.annotation_path)
# Do LOESS correction if specified
if self.LOESS:
self.transit_message("Performing LOESS Correction")
for j in range(K):
data[j] = stat_tools.loess_correction(position, data[j])
# Get Gene objects for each condition
G_A1 = tnseq_tools.Genes([],
self.annotation_path,
data=data[:Na1],
position=position,
nterm=self.NTerminus,
cterm=self.CTerminus)
G_B1 = tnseq_tools.Genes([],
self.annotation_path,
data=data[Na1:(Na1 + Nb1)],
position=position,
nterm=self.NTerminus,
cterm=self.CTerminus)
G_A2 = tnseq_tools.Genes([],
self.annotation_path,
data=data[(Na1 + Nb1):(Na1 + Nb1 + Na2)],
position=position,
nterm=self.NTerminus,
cterm=self.CTerminus)
G_B2 = tnseq_tools.Genes([],
self.annotation_path,
data=data[(Na1 + Nb1 + Na2):],
position=position,
nterm=self.NTerminus,
cterm=self.CTerminus)
means_list_a1 = []
means_list_b1 = []
means_list_a2 = []
means_list_b2 = []
var_list_a1 = []
var_list_a2 = []
var_list_b1 = []
var_list_b2 = []
# Base priors on empirical observations across genes.
for gene in sorted(G_A1):
if gene.n > 1:
A1_data = G_A1[gene.orf].reads.flatten()
B1_data = G_B1[gene.orf].reads.flatten()
A2_data = G_A2[gene.orf].reads.flatten()
B2_data = G_B2[gene.orf].reads.flatten()
means_list_a1.append(numpy.mean(A1_data))
var_list_a1.append(numpy.var(A1_data))
means_list_b1.append(numpy.mean(B1_data))
var_list_b1.append(numpy.var(B1_data))
means_list_a2.append(numpy.mean(A2_data))
var_list_a2.append(numpy.var(A2_data))
means_list_b2.append(numpy.mean(B2_data))
var_list_b2.append(numpy.var(B2_data))
# Priors
mu0_A1 = scipy.stats.trim_mean(means_list_a1, 0.01)
mu0_B1 = scipy.stats.trim_mean(means_list_b1, 0.01)
mu0_A2 = scipy.stats.trim_mean(means_list_a2, 0.01)
mu0_B2 = scipy.stats.trim_mean(means_list_b2, 0.01)
s20_A1 = scipy.stats.trim_mean(var_list_a1, 0.01)
s20_B1 = scipy.stats.trim_mean(var_list_b1, 0.01)
s20_A2 = scipy.stats.trim_mean(var_list_a2, 0.01)
s20_B2 = scipy.stats.trim_mean(var_list_b2, 0.01)
k0 = 1.0
nu0 = 1.0
data = []
postprob = []
count = 0
N = len(G_A1)
self.progress_range(N)
# Perform actual analysis
for gene in G_A1:
# If there is some data
if gene.n > 0:
A1_data = G_A1[gene.orf].reads.flatten()
B1_data = G_B1[gene.orf].reads.flatten()
A2_data = G_A2[gene.orf].reads.flatten()
B2_data = G_B2[gene.orf].reads.flatten()
# Time-1 Time-2
#
# Strain-A A C
#
# Strain-B B D
try:
muA1_post, varA1_post = stat_tools.sample_trunc_norm_post(
A1_data, self.samples, mu0_A1, s20_A1, k0, nu0)
muB1_post, varB1_post = stat_tools.sample_trunc_norm_post(
B1_data, self.samples, mu0_B1, s20_B1, k0, nu0)
muA2_post, varA2_post = stat_tools.sample_trunc_norm_post(
A2_data, self.samples, mu0_A2, s20_A2, k0, nu0)
muB2_post, varB2_post = stat_tools.sample_trunc_norm_post(
B2_data, self.samples, mu0_B2, s20_B2, k0, nu0)
except Exception as e:
muA1_post = varA1_post = numpy.ones(self.samples)
muB1_post = varB1_post = numpy.ones(self.samples)
muA2_post = varA2_post = numpy.ones(self.samples)
muB2_post = varB2_post = numpy.ones(self.samples)
logFC_A_post = numpy.log2(muA2_post / muA1_post)
logFC_B_post = numpy.log2(muB2_post / muB1_post)
delta_logFC_post = logFC_B_post - logFC_A_post
alpha = 0.05
# Get Bounds of the HDI
l_logFC_A, u_logFC_A = stat_tools.HDI_from_MCMC(
logFC_A_post, 1 - alpha)
l_logFC_B, u_logFC_B = stat_tools.HDI_from_MCMC(
logFC_B_post, 1 - alpha)
l_delta_logFC, u_delta_logFC = stat_tools.HDI_from_MCMC(
delta_logFC_post, 1 - alpha)
mean_logFC_A = numpy.mean(logFC_A_post)
mean_logFC_B = numpy.mean(logFC_B_post)
mean_delta_logFC = numpy.mean(delta_logFC_post)
# Is HDI significantly different than ROPE? (i.e. no overlap)
not_HDI_overlap_bit = l_delta_logFC > self.rope or u_delta_logFC < -self.rope
# Probability of posterior overlaping with ROPE
probROPE = numpy.mean(
numpy.logical_and(delta_logFC_post >= 0.0 - self.rope,
delta_logFC_post <= 0.0 + self.rope))
# If there is no data, assume empty defaults
else:
A1_data = [0, 0]
B1_data = [0, 0]
A2_data = [0, 0]
B2_data = [0, 0]
muA1_post = varA1_post = numpy.ones(self.samples)
muB1_post = varB1_post = numpy.ones(self.samples)
muA2_post = varA2_post = numpy.ones(self.samples)
muB2_post = varB2_post = numpy.ones(self.samples)
logFC_A_post = numpy.log2(muA2_post / muA1_post)
logFC_B_post = numpy.log2(muB2_post / muB1_post)
delta_logFC_post = logFC_B_post - logFC_A_post
mean_logFC_A = 0
mean_logFC_B = 0
mean_delta_logFC = 0
l_logFC_A = 0
u_logFC_A = 0
l_logFC_B = 0
u_logFC_B = 0
l_delta_logFC = 0
u_delta_logFC = 0
probROPE = 1.0
if numpy.isnan(l_logFC_A):
l_logFC_A = -10
u_logFC_A = 10
if numpy.isnan(l_logFC_B):
l_logFC_B = -10
u_logFC_B = 10
if numpy.isnan(l_delta_logFC):
l_delta_logFC = -10
u_delta_logFC = 10
postprob.append(probROPE)
data.append((gene.orf, gene.name, gene.n, numpy.mean(muA1_post),
numpy.mean(muA2_post), numpy.mean(muB1_post),
numpy.mean(muB2_post), mean_logFC_A, mean_logFC_B,
mean_delta_logFC, l_delta_logFC, u_delta_logFC,
probROPE, not_HDI_overlap_bit))
text = "Running GI Method... %2.0f%%" % (100.0 * (count + 1) / N)
self.progress_update(text, count)
self.transit_message_inplace("Running Export Method... %1.1f%%" %
(100.0 * count / (N - 1)))
count += 1
# for HDI, maybe I should sort on abs(mean_delta_logFC); however, need to sort by prob to calculate BFDR
probcol = -2 # probROPEs
data.sort(key=lambda x: x[probcol])
sortedprobs = numpy.array([x[probcol] for x in data])
# BFDR method: Newton M.A., Noueiry A., Sarkar D., Ahlquist P. (2004). Detecting differential gene expression with a semiparametric hierarchical mixture method. Biostatistics, 5:155–176.
if self.signif == "BFDR":
sortedprobs = numpy.array(sortedprobs)
#sortedprobs.sort() # why, since already sorted?
bfdr = numpy.cumsum(sortedprobs) / numpy.arange(
1,
len(sortedprobs) + 1)
adjusted_prob = bfdr # should be same order as sorted above by probROPE
adjusted_label = "BFDR"
elif self.signif == "FWER":
fwer = stat_tools.FWER_Bayes(sortedprobs)
#fwer.sort() # should not need this if monotonic
adjusted_prob = fwer
adjusted_label = "FWER"
# If not using adjustment for classification, sort correctly
else:
adjusted_prob = sortedprobs
adjusted_label = "un"
# should I stable-sort by overlap_bit?
# sorted_index = numpy.argsort([d[-1] for d in data])[::-1][:len(data)]
# adjusted_prob = [adjusted_prob[ii] for ii in sorted_index]
# data = [data[ii] for ii in sorted_index]
# Print(output)
if self.wxobj:
members = sorted([
attr for attr in dir(self) if not callable(getattr(self, attr))
and not attr.startswith("__")
])
memberstr = ""
for m in members:
memberstr += "%s = %s, " % (m, getattr(self, m))
self.output.write(
"#GUI with: norm=%s, samples=%s, includeZeros=%s, output=%s\n"
% (self.normalization, self.samples, self.includeZeros,
self.output.name.encode('utf-8')))
else:
self.output.write("#Console: python3 %s\n" % " ".join(sys.argv))
now = str(datetime.datetime.now())
now = now[:now.rfind('.')]
self.output.write("#Date: " + now + "\n")
#self.output.write("#Runtime: %s s\n" % (time.time() - start_time))
self.output.write("#Control Data-A: %s\n" %
(",".join(self.ctrldataA).encode('utf-8')))
self.output.write("#Control Data-B: %s\n" %
(",".join(self.ctrldataB).encode('utf-8')))
self.output.write("#Experimental Data-A: %s\n" %
(",".join(self.expdataA).encode('utf-8')))
self.output.write("#Experimental Data-B: %s\n" %
(",".join(self.expdataB).encode('utf-8')))
self.output.write("#Annotation path: %s\n" %
(self.annotation_path.encode('utf-8')))
self.output.write("#ROPE=%s, method for significance=%s\n" %
(self.rope, self.signif))
#self.output.write("#%s\n" % "\t".join(columns))
if self.signif == "HDI":
self.output.write(
"#Significant interactions are those genes whose delta-logFC HDI does not overlap the ROPE\n"
)
elif self.signif in "prob BDFR FWER":
self.output.write(
"#Significant interactions are those whose %s-adjusted probability of the delta-logFC falling within ROPE is < 0.05.\n"
% (adjusted_label))
# Write column names (redundant with self.columns)
self.output.write(
"#ORF\tName\tNumber of TA Sites\tMean count (Strain A Condition 1)\tMean count (Strain A Condition 2)\tMean count (Strain B Condition 1)\tMean count (Strain B Condition 2)\tMean logFC (Strain A)\tMean logFC (Strain B) \tMean delta logFC\tLower Bound delta logFC\tUpper Bound delta logFC\tIs HDI outside ROPE?\tProb. of delta-logFC being within ROPE\t%s-Adjusted Probability\tType of Interaction\n"
% adjusted_label)
# Write gene results
for i, row in enumerate(data):
#1 2 3 4 5 6 7 8 9 10 11 12 13 14
orf, name, n, mean_muA1_post, mean_muA2_post, mean_muB1_post, mean_muB2_post, mean_logFC_A, mean_logFC_B, mean_delta_logFC, l_delta_logFC, u_delta_logFC, probROPE, not_HDI_overlap_bit = row
interaction = self.classify_interaction(mean_delta_logFC,
mean_logFC_B, mean_logFC_A)
type_of_interaction = "No Interaction"
if self.signif in "prob BFDR FWER" and adjusted_prob[i] < 0.05:
type_of_interaction = interaction
if self.signif == "HDI" and not_HDI_overlap_bit:
type_of_interaction = interaction
new_row = tuple(
list(row[:-2]) + [
not_HDI_overlap_bit, probROPE, adjusted_prob[i],
type_of_interaction
])
self.output.write(
"%s\t%s\t%d\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%1.2f\t%s\t%1.8f\t%1.8f\t%s\n"
% new_row)
self.transit_message("Adding File: %s" % (self.output.name))
self.add_file(filetype="GI")
self.finish()
self.transit_message("Finished Genetic Interactions Method")