def pca(data, algorithm='eig'):
"""pca(data) -> mean, pcs, norm_pcs, variances, positions, norm_positions
Perform Principal Components Analysis on a set of n data points in k
dimensions. The data array must be of shape (n, k).
This function returns the transformed position of each data point along
with the amount of variance captured by each principal component.
The optional algorithm parameter can be either 'svd' to perform PCA with
the singular value decomposition, or 'eig' to use a symmetric eigenvalue
decomposition. Empirically, eig is faster on the datasets I have tested.
"""
if False: #scaling is disabled
for i in xrange(0, len(data[0])):
freqs = []
for j in xrange(0, len(data)):
freqs.append(data[j][i])
meanFreq = float(sum(freqs)) / len(freqs)
sdFreq = stdDev(freqs, meanFreq)
for j in xrange(0, len(data)):
data[j][i] -= meanFreq
data[j][i] /= sdFreq
data = numpy.asarray(data)
mean = data.mean(axis=0)
centered = data - mean
if algorithm == 'eig':
pcs, variances, stds, positions, norm_positions = _pca_eig(centered)
elif algorithm == 'svd':
pcs, variances, stds, positions, norm_positions = _pca_svd(centered)
else:
raise RuntimeError('Algorithm %s not known.' % algorithm)
sumVariances = sum(variances)
for i in xrange(0, len(variances)):
variances[i] /= sumVariances
return positions, variances
def pca(data, algorithm='eig'):
"""pca(data) -> mean, pcs, norm_pcs, variances, positions, norm_positions
Perform Principal Components Analysis on a set of n data points in k
dimensions. The data array must be of shape (n, k).
This function returns the transformed position of each data point along
with the amount of variance captured by each principal component.
The optional algorithm parameter can be either 'svd' to perform PCA with
the singular value decomposition, or 'eig' to use a symmetric eigenvalue
decomposition. Empirically, eig is faster on the datasets I have tested.
"""
if False: #scaling is disabled
for i in xrange(0, len(data[0])):
freqs = []
for j in xrange(0, len(data)):
freqs.append(data[j][i])
meanFreq = float(sum(freqs)) / len(freqs)
sdFreq = stdDev(freqs, meanFreq)
for j in xrange(0, len(data)):
data[j][i] -= meanFreq
data[j][i] /= sdFreq
data = numpy.asarray(data)
mean = data.mean(axis = 0)
centered = data - mean
if algorithm=='eig':
pcs, variances, stds, positions, norm_positions = _pca_eig(centered)
elif algorithm=='svd':
pcs, variances, stds, positions, norm_positions = _pca_svd(centered)
else:
raise RuntimeError('Algorithm %s not known.'%algorithm)
sumVariances = sum(variances)
for i in xrange(0, len(variances)):
variances[i] /= sumVariances
return positions, variances
def run(self, statTest, effectSizeMeasure, profile, progress=None):
self.results.data = []
self.results.test = statTest.name
self.results.profile = profile
if progress == 'Verbose':
print ' Processing feature:'
index = 0
for feature in profile.getFeatures():
if progress == 'Verbose':
print ' ' + feature
elif progress != None:
if progress.wasCanceled():
self.results.data = []
return
index += 1
progress.setValue(index)
seqCount = profile.getActiveFeatureCounts(feature)
parentCount = profile.getActiveParentCounts(feature)
data = profile.getActiveFeatureProportions(feature)
pValue, note = statTest.hypothesisTest(data)
effectSize = effectSizeMeasure.run(data)
row = [feature, float(pValue), float(pValue), effectSize, note]
for i in xrange(0, len(seqCount)):
propGroup = []
for j in xrange(0, len(seqCount[i])):
sc = seqCount[i][j]
pc = parentCount[i][j]
if pc > 0:
propGroup.append(sc * 100.0 / pc)
else:
propGroup.append(0.0)
meanGroup = mean(propGroup)
row.append(meanGroup)
row.append(stdDev(propGroup, meanGroup))
for i in xrange(0, len(seqCount)):
for j in xrange(0, len(seqCount[i])):
sc = seqCount[i][j]
pc = parentCount[i][j]
row.append(sc)
row.append(pc)
if pc > 0:
row.append(sc * 100.0 / pc)
else:
row.append(0.0)
self.results.data.append(row)
headingsSampleStats = []
for i in xrange(0, len(profile.activeSamplesInGroups)):
for sampleName in profile.activeSamplesInGroups[i]:
headingsSampleStats.append(sampleName)
headingsSampleStats.append(sampleName + ': parent seq. count')
headingsSampleStats.append(sampleName + ': rel. freq. (%)')
self.results.createTableHeadings(profile.activeGroupNames,
headingsSampleStats)
if len(self.results.data) >= 1:
# sort results according to p-values
self.results.data = TableHelper.SortTable(
self.results.data, [self.results.dataHeadings['pValues']])
if progress != None and progress != 'Verbose':
index += 1
progress.setValue(index)
def run(self, test, signLevel, statsResults, trials, bootstrapRep,
progress):
tableData = []
index = 0
for row in statsResults:
feature = row[0]
seq1 = row[1]
seq2 = row[2]
parentSeq1 = row[3]
parentSeq2 = row[4]
p1 = float(seq1) / parentSeq1
p2 = float(seq2) / parentSeq2
powerList = []
powerListLess5 = []
powerListGreater5 = []
for trial in xrange(0, trials):
if progress != '':
index += 1
progress.setValue(index)
progress.setLabelText(feature + ' - Trial = ' + str(trial))
power = 0
processedReplicates = 0
for dummy in xrange(0, bootstrapRep):
c1 = 0
c2 = 0
for dummy in xrange(0, parentSeq1):
rnd = random.random()
if rnd <= p1:
c1 += 1
for dummy in xrange(0, parentSeq2):
rnd = random.random()
if rnd <= p2:
c2 += 1
if c1 == 0 and c2 == 0:
# This is a special case that many hypothesis test will not handle correctly
# so we just ignore it. This will have little effect on the calculated power
# of a test.
continue
processedReplicates += 1
pValueOneSided, pValueTwoSided = test.hypothesisTest(
c1, c2, parentSeq1, parentSeq2)
if pValueTwoSided < signLevel:
power += 1
if processedReplicates > 0:
if min([seq1, seq2]) <= 5:
powerListLess5.append(
float(power) / processedReplicates)
else:
powerListGreater5.append(
float(power) / processedReplicates)
powerList.append(float(power) / processedReplicates)
row = []
row.append(feature)
row.append(seq1)
row.append(seq2)
row.append(parentSeq1)
row.append(parentSeq2)
row.append(float(seq1) / parentSeq1)
row.append(float(seq2) / parentSeq2)
row.append(mean(powerList))
row.append(stdDev(powerList))
if math.isnan(mean(powerListLess5)):
row.append('')
else:
row.append(mean(powerListLess5))
if math.isnan(stdDev(powerListLess5)):
row.append('')
else:
row.append(stdDev(powerListLess5))
if math.isnan(mean(powerListGreater5)):
row.append('')
else:
row.append(mean(powerListGreater5))
if math.isnan(stdDev(powerListGreater5)):
row.append('')
else:
row.append(stdDev(powerListGreater5))
tableData.append(row)
return tableData
def run(self, statTest, effectSizeMeasure, profile, progress = None): self.results.data = [] self.results.test = statTest.name self.results.profile = profile if progress == 'Verbose': print ' Processing feature:' index = 0 for feature in profile.getFeatures(): if progress == 'Verbose': print ' ' + feature elif progress != None: if progress.wasCanceled(): self.results.data = [] return index += 1 progress.setValue(index) seqCount = profile.getActiveFeatureCounts(feature) parentCount = profile.getActiveParentCounts(feature) data = profile.getActiveFeatureProportions(feature) pValue, note = statTest.hypothesisTest(data) effectSize = effectSizeMeasure.run(data) row = [feature, float(pValue), float(pValue), effectSize, note] for i in xrange(0, len(seqCount)): propGroup = [] for j in xrange(0, len(seqCount[i])): propGroup.append(seqCount[i][j] * 100.0 / parentCount[i][j]) meanGroup = mean(propGroup) row.append(meanGroup) row.append(stdDev(propGroup, meanGroup)) for i in xrange(0, len(seqCount)): for j in xrange(0, len(seqCount[i])): row.append(seqCount[i][j]) row.append(parentCount[i][j]) row.append(seqCount[i][j] * 100.0 / parentCount[i][j]) self.results.data.append(row) headingsSampleStats = [] for i in xrange(0, len(profile.activeSamplesInGroups)): for sampleName in profile.activeSamplesInGroups[i]: headingsSampleStats.append(sampleName) headingsSampleStats.append(sampleName + ': parent seq. count') headingsSampleStats.append(sampleName + ': rel. freq. (%)') self.results.createTableHeadings(profile.activeGroupNames, headingsSampleStats) if len(self.results.data) >= 1: # sort results according to p-values self.results.data = TableHelper.SortTable(self.results.data, [self.results.dataHeadings['pValues']]) if progress != None and progress != 'Verbose': index += 1 progress.setValue(index)
def plot(self, profile, statsResults):
if len(profile.profileDict) <= 0:
self.emptyAxis()
return
if len(profile.profileDict) > 10000:
QtGui.QApplication.instance().setOverrideCursor(
QtGui.QCursor(QtCore.Qt.ArrowCursor))
reply = QtGui.QMessageBox.question(
self, 'Continue?', 'Profile contains ' +
str(len(profile.profileDict)) + ' features. ' +
'It may take several seconds to generate this plot. Exploring the data at a higher hierarchy level is recommended. '
+ 'Do you wish to continue?', QtGui.QMessageBox.Yes,
QtGui.QMessageBox.No)
QtGui.QApplication.instance().restoreOverrideCursor()
if reply == QtGui.QMessageBox.No:
self.emptyAxis()
return
# *** Colour of plot elements
axesColour = str(self.preferences['Axes colour'].name())
group1Colour = str(
self.preferences['Group colours'][profile.groupName1].name())
group2Colour = str(
self.preferences['Group colours'][profile.groupName2].name())
# *** Set sample names
self.groupName1 = profile.groupName1
self.groupName2 = profile.groupName2
# *** Create lists for each quantity of interest and calculate spread of data
groupData1, groupData2 = profile.getFeatureProportionsAll()
features = profile.getFeatures()
field1 = []
field2 = []
xSpread = []
ySpread = []
for i in xrange(0, len(groupData1)):
mean1 = mean(groupData1[i])
mean2 = mean(groupData2[i])
field1.append(mean1)
field2.append(mean2)
if self.spreadMethod == 'standard deviation':
xSpread.append([
max(mean1 - stdDev(groupData1[i], mean1), 0),
min(mean1 + stdDev(groupData1[i], mean1), 100)
])
ySpread.append([
max(mean2 - stdDev(groupData2[i], mean2), 0),
min(mean2 + stdDev(groupData2[i], mean2), 100)
])
elif self.spreadMethod == '2 * standard deviation':
xSpread.append([
max(mean1 - 2 * stdDev(groupData1[i], mean1), 0),
min(mean1 + 2 * stdDev(groupData1[i], mean1), 100)
])
ySpread.append([
max(mean2 - 2 * stdDev(groupData2[i], mean2), 0),
min(mean2 + 2 * stdDev(groupData2[i], mean2), 100)
])
elif self.spreadMethod == '25th and 75th percentile':
spread1 = mquantiles(groupData1[i], prob=[0.25, 0.75])
spread2 = mquantiles(groupData2[i], prob=[0.25, 0.75])
xSpread.append([max(spread1[0], 0), min(spread1[1], 100)])
ySpread.append([max(spread2[0], 0), min(spread2[1], 100)])
elif self.spreadMethod == '9th and 91st percentile':
spread1 = mquantiles(groupData1[i], prob=[0.09, 0.91])
spread2 = mquantiles(groupData2[i], prob=[0.09, 0.91])
xSpread.append([max(spread1[0], 0), min(spread1[1], 100)])
ySpread.append([max(spread2[0], 0), min(spread2[1], 100)])
elif self.spreadMethod == '2nd and 98th percentile':
spread1 = mquantiles(groupData1[i], prob=[0.02, 0.98])
spread2 = mquantiles(groupData2[i], prob=[0.02, 0.98])
xSpread.append([max(spread1[0], 0), min(spread1[1], 100)])
ySpread.append([max(spread2[0], 0), min(spread2[1], 100)])
elif self.spreadMethod == 'minimum and maximum':
xSpread.append([max(groupData1[i]), min(groupData1[i])])
ySpread.append([max(groupData2[i]), min(groupData2[i])])
# *** Set figure size
self.fig.clear()
self.fig.set_size_inches(self.figWidth, self.figHeight)
if self.bShowHistograms:
histogramSizeX = self.histogramSize / self.figWidth
histogramSizeY = self.histogramSize / self.figHeight
else:
histogramSizeX = 0.0
histogramSizeY = 0.0
padding = 0.1 # inches
xOffsetFigSpace = (0.4 + padding) / self.figWidth
yOffsetFigSpace = (0.3 + padding) / self.figHeight
axesScatter = self.fig.add_axes([
xOffsetFigSpace, yOffsetFigSpace, 1.0 - xOffsetFigSpace -
histogramSizeX - (2 * padding) / self.figWidth, 1.0 -
yOffsetFigSpace - histogramSizeY - (2 * padding) / self.figHeight
])
if self.bShowHistograms:
axesTopHistogram = self.fig.add_axes([
xOffsetFigSpace,
1.0 - histogramSizeY - padding / self.figHeight,
1.0 - xOffsetFigSpace - histogramSizeX -
(2 * padding) / self.figWidth, histogramSizeY
])
axesRightHistogram = self.fig.add_axes([
1.0 - histogramSizeX - padding / self.figWidth,
yOffsetFigSpace, histogramSizeX, 1.0 - yOffsetFigSpace -
histogramSizeY - (2 * padding) / self.figHeight
])
# *** Handle mouse events
tooltips = []
for i in xrange(0, len(field1)):
tooltip = features[i] + '\n\n'
tooltip += (self.groupName1 +
' mean proportion: %.3f' % field1[i]) + '\n'
tooltip += (self.groupName2 +
' mean proportion: %.3f' % field2[i]) + '\n\n'
tooltip += 'Difference between mean proportions (%): ' + (
'%.3f' % (field1[i] - field2[i])) + '\n'
if field2[i] != 0:
tooltip += 'Ratio of mean proportions: %.3f' % (field1[i] /
field2[i])
else:
tooltip += 'Ratio of mean proportions: undefined'
if statsResults.profile != None:
pValue = statsResults.getFeatureStatisticAsStr(
features[i], 'pValues')
pValueCorrected = statsResults.getFeatureStatisticAsStr(
features[i], 'pValuesCorrected')
tooltip += '\n\n'
tooltip += 'p-value: ' + pValue + '\n'
tooltip += 'Corrected p-value: ' + pValueCorrected
tooltips.append(tooltip)
self.plotEventHandler = PlotEventHandler(field1, field2, tooltips)
self.mouseEventCallback(self.plotEventHandler)
# *** Calculate R^2 value
slope, intercept, r_value, p_value, std_err = linregress(
field1, field2)
# *** Plot data
# set visual properties of all points
colours = []
highlightedField1 = []
highlightedField2 = []
highlighColours = []
for i in xrange(0, len(field1)):
if field1[i] > field2[i]:
colours.append(group1Colour)
else:
colours.append(group2Colour)
if features[i] in self.preferences['Highlighted group features']:
highlightedField1.append(field1[i])
highlightedField2.append(field2[i])
highlighColours.append(colours[i])
# scatter plot
axesScatter.scatter(field1,
field2,
c=colours,
s=self.markerSize,
zorder=5)
if len(highlightedField1) > 0:
axesScatter.scatter(highlightedField1,
highlightedField2,
c=highlighColours,
s=self.markerSize,
edgecolors='red',
linewidth=2,
zorder=10)
# plot CIs
if self.spreadMethod != 'None':
xlist = []
ylist = []
for i in xrange(0, len(field1)):
# horizontal CIs
xlist.append(xSpread[i][0])
xlist.append(xSpread[i][1])
xlist.append(None)
ylist.append(field2[i])
ylist.append(field2[i])
ylist.append(None)
# vertical CIs
xlist.append(field1[i])
xlist.append(field1[i])
xlist.append(None)
ylist.append(ySpread[i][0])
ylist.append(ySpread[i][1])
ylist.append(None)
axesScatter.plot(xlist,
ylist,
'-',
color='gray',
antialiased=False)
# plot y=x line
maxProportion = max(max(field1), max(field2)) * 1.05
axesScatter.plot([0, maxProportion], [0, maxProportion],
color=axesColour,
linestyle='dashed',
marker='',
zorder=1)
axesScatter.set_xlabel(self.groupName1 + ' (%)')
axesScatter.set_ylabel(self.groupName2 + ' (%)')
if self.bShowR2:
axesScatter.text(0.02,
0.98,
r'R$^2$ = ' + ('%0.3f' % r_value**2),
horizontalalignment='left',
verticalalignment='top',
transform=axesScatter.transAxes)
axesScatter.set_xlim(0, maxProportion)
axesScatter.set_ylim(0, maxProportion)
# *** Prettify scatter plot
for line in axesScatter.yaxis.get_ticklines():
line.set_color(axesColour)
for line in axesScatter.xaxis.get_ticklines():
line.set_color(axesColour)
for loc, spine in axesScatter.spines.iteritems():
spine.set_color(axesColour)
# plot histograms
if not self.bShowHistograms:
for a in axesScatter.yaxis.majorTicks:
a.tick1On = True
a.tick2On = False
for a in axesScatter.xaxis.majorTicks:
a.tick1On = True
a.tick2On = False
for line in axesScatter.yaxis.get_ticklines():
line.set_color(axesColour)
for line in axesScatter.xaxis.get_ticklines():
line.set_color(axesColour)
for loc, spine in axesScatter.spines.iteritems():
if loc in ['right', 'top']:
spine.set_color('none')
else:
spine.set_color(axesColour)
else: # show histograms
# plot top histogram
axesTopHistogram.xaxis.set_major_formatter(NullFormatter())
pdf, bins, patches = axesTopHistogram.hist(field1,
bins=self.numBins,
facecolor=group1Colour)
axesTopHistogram.set_xlim(axesScatter.get_xlim())
axesTopHistogram.set_yticks([0, max(pdf)])
axesTopHistogram.set_ylim([0, max(pdf) * 1.05])
# plot right histogram
axesRightHistogram.yaxis.set_major_formatter(NullFormatter())
pdf, bins, patches = axesRightHistogram.hist(
field2,
bins=self.numBins,
orientation='horizontal',
facecolor=group2Colour)
axesRightHistogram.set_ylim(axesScatter.get_ylim())
axesRightHistogram.set_xticks([0, max(pdf)])
axesRightHistogram.set_xlim([0, max(pdf) * 1.05])
# *** Prettify histogram plot
for a in axesTopHistogram.yaxis.majorTicks:
a.tick1On = True
a.tick2On = False
for a in axesTopHistogram.xaxis.majorTicks:
a.tick1On = True
a.tick2On = False
for line in axesTopHistogram.yaxis.get_ticklines():
line.set_color(axesColour)
for line in axesTopHistogram.xaxis.get_ticklines():
line.set_color(axesColour)
for loc, spine in axesTopHistogram.spines.iteritems():
if loc in ['right', 'top']:
spine.set_color('none')
else:
spine.set_color(axesColour)
for a in axesRightHistogram.yaxis.majorTicks:
a.tick1On = True
a.tick2On = False
for a in axesRightHistogram.xaxis.majorTicks:
a.tick1On = True
a.tick2On = False
for line in axesRightHistogram.yaxis.get_ticklines():
line.set_color(axesColour)
for line in axesRightHistogram.xaxis.get_ticklines():
line.set_color(axesColour)
for loc, spine in axesRightHistogram.spines.iteritems():
if loc in ['right', 'top']:
spine.set_color('none')
else:
spine.set_color(axesColour)
self.updateGeometry()
self.draw()
results = [
coverageListDP, coverageListDPCC, coverageListNW, coverageListWoolf,
coverageListGart, coverageListRP
]
lengths = [
ciLengthDP, ciLengthDPCC, ciLengthNW, ciLengthWoolf, ciLengthGart,
ciLengthRP
]
methodNames = [
'DP: Asymptotic', 'DP: Asymptotic-CC', 'Newcombe-Wilson', 'Woolf', 'Gart',
'RP: Asympototic'
]
for i in xrange(0, len(results)):
coverageMeanStr = '%.2f' % mean(results[i])
coverageSdStr = '%.2f' % stdDev(results[i])
coverageMinStr = '%.2f' % min(results[i])
coverageMaxStr = '%.2f' % max(results[i])
lengthMeanStr = '%.2f' % mean(lengths[i])
lengthSdStr = '%.2f' % stdDev(lengths[i])
fout.write(methodNames[i] + '\n')
fout.write(coverageMeanStr + '+/-' + coverageSdStr + '[' + coverageMinStr +
';' + coverageMaxStr + ']\n')
fout.write(lengthMeanStr + '+/-' + lengthSdStr + '\n')
fout.write('\n')
fout.close()
def run(self, statTest, testType, confIntervMethod, coverage, profile, progress = None): self.results.test = statTest.name self.results.testType = testType self.results.alpha = 1.0 - coverage self.results.confIntervMethod = confIntervMethod self.results.profile = profile if progress == 'Verbose': print ' Processing feature:' self.results.data = [] index = 0 # calculate statistics seqsGroup1 = [] seqsGroup2 = [] parentSeqsGroup1 = [] parentSeqsGroup2 = [] pValues = [] lowerCIs = [] upperCIs = [] effectSizes = [] notes = [] if statTest.bSingleFeatureInterface: # process features one at a time for feature in profile.getFeatures(): if progress == 'Verbose': print ' ' + feature elif progress != None: if progress.wasCanceled(): self.results.data = [] return index += 1 progress.setValue(index) # get statistics seqGroup1, seqGroup2 = profile.getFeatureCounts(feature) parentSeqGroup1, parentSeqGroup2= profile.getParentFeatureCounts(feature) results = statTest.run(seqGroup1, seqGroup2, parentSeqGroup1, parentSeqGroup2, confIntervMethod, coverage) pValueOneSided, pValueTwoSided, lowerCI, upperCI, effectSize, note = results if testType == 'One-sided': pValue = pValueOneSided elif testType == 'Two-sided': pValue = pValueTwoSided else: print 'Error: Unknown test type.' # record results seqsGroup1.append(seqGroup1) seqsGroup2.append(seqGroup2) parentSeqsGroup1.append(parentSeqGroup1) parentSeqsGroup2.append(parentSeqGroup2) pValues.append(pValue) lowerCIs.append(lowerCI) upperCIs.append(upperCI) effectSizes.append(effectSize) notes.append(note) if progress != None and progress != 'Verbose': index += 1 progress.setValue(index) else: # process all features at once seqsGroup1, seqsGroup2 = profile.getFeatureCountsAll() parentSeqsGroup1, parentSeqsGroup2= profile.getParentFeatureCountsAll() pValuesOneSided, pValuesTwoSided, lowerCIs, upperCIs, effectSizes, notes = statTest.runAll(seqsGroup1, seqsGroup2, parentSeqsGroup1, parentSeqsGroup2, confIntervMethod, coverage, progress) if progress == 'Verbose': print ' Processing all features...' elif progress != None and progress.wasCanceled(): self.results.data = [] return if testType == 'One-sided': pValues = pValuesOneSided elif testType == 'Two-sided': pValues = pValuesTwoSided else: print 'Error: Unknown test type.' # record statistics features = profile.getFeatures() for i in xrange(0, len(features)): propGroup1 = [] for j in xrange(0, len(seqsGroup1[i])): sg1 = seqsGroup1[i][j] psg1 = parentSeqsGroup1[i][j] if psg1 > 0: propGroup1.append( sg1 * 100.0 / psg1 ) else: propGroup1.append( 0.0 ) propGroup2 = [] for j in xrange(0, len(seqsGroup2[i])): sg2 = seqsGroup2[i][j] psg2 = parentSeqsGroup2[i][j] if psg2 > 0: propGroup2.append( sg2 * 100.0 / psg2 ) else: propGroup2.append( 0.0 ) meanGroup1 = mean(propGroup1) meanGroup2 = mean(propGroup2) row = [features[i], meanGroup1, stdDev(propGroup1, meanGroup1), meanGroup2, stdDev(propGroup2, meanGroup2), float(pValues[i]),float(pValues[i]),float(effectSizes[i]), float(lowerCIs[i]),float(upperCIs[i]), notes[i]] for j in xrange(0, len(seqsGroup1[i])): row.append(seqsGroup1[i][j]) row.append(parentSeqsGroup1[i][j]) if parentSeqsGroup1[i][j] > 0: row.append(seqsGroup1[i][j] * 100.0 / parentSeqsGroup1[i][j]) else: row.append(0.0) for j in xrange(0, len(seqsGroup2[i])): row.append(seqsGroup2[i][j]) row.append(parentSeqsGroup2[i][j]) if parentSeqsGroup2[i][j] > 0: row.append(seqsGroup2[i][j] * 100.0 / parentSeqsGroup2[i][j]) else: row.append(0.0) self.results.data.append(row) headingsSampleStats = [] for sampleName in (profile.samplesInGroup1 + profile.samplesInGroup2): headingsSampleStats.append(sampleName) headingsSampleStats.append(sampleName + ': parent seq. count') headingsSampleStats.append(sampleName + ': rel. freq. (%)') self.results.createTableHeadings(profile.groupName1, profile.groupName2, headingsSampleStats) # sort results according to p-values if len(self.results.data) >= 1: self.results.data = TableHelper.SortTable(self.results.data, [self.results.dataHeadings['pValues']])
def run(self, confIntervMethod, coverage, tables, trials, bootstrapRep,
progress):
tableData = []
index = 0
for row in tables:
feature = row[0]
seq1 = row[1]
seq2 = row[2]
parentSeq1 = row[3]
parentSeq2 = row[4]
lowerCI, upperCI, obsEffectSize = confIntervMethod.run(
seq1, seq2, parentSeq1, parentSeq2, coverage)
p1 = float(seq1) / parentSeq1
p2 = float(seq2) / parentSeq2
coverageList = []
coverageListLess5 = []
coverageListGreater5 = []
for trial in xrange(0, trials):
if progress != '':
index += 1
progress.setValue(index)
progress.setLabelText(feature + ' - Trial = ' + str(trial))
containedRep = 0
for dummy in xrange(0, bootstrapRep):
c1 = binomial(parentSeq1, p1)
c2 = binomial(parentSeq2, p2)
lowerCI, upperCI, effectSize = confIntervMethod.run(
c1, c2, parentSeq1, parentSeq2, coverage)
if obsEffectSize >= lowerCI and obsEffectSize <= upperCI:
containedRep += 1
if min([seq1, seq2]) <= 5:
coverageListLess5.append(
float(containedRep) / bootstrapRep)
else:
coverageListGreater5.append(
float(containedRep) / bootstrapRep)
coverageList.append(float(containedRep) / bootstrapRep)
row = []
row.append(feature)
row.append(seq1)
row.append(seq2)
row.append(parentSeq1)
row.append(parentSeq2)
row.append(float(seq1) / parentSeq1)
row.append(float(seq2) / parentSeq2)
row.append(mean(coverageList))
row.append(stdDev(coverageList))
if math.isnan(mean(coverageListLess5)):
row.append('')
else:
row.append(mean(coverageListLess5))
if math.isnan(stdDev(coverageListLess5)):
row.append('')
else:
row.append(stdDev(coverageListLess5))
if math.isnan(mean(coverageListGreater5)):
row.append('')
else:
row.append(mean(coverageListGreater5))
if math.isnan(stdDev(coverageListGreater5)):
row.append('')
else:
row.append(stdDev(coverageListGreater5))
tableData.append(row)
return tableData
def run(self, test, signLevel, statsResults, trials, bootstrapRep, progress):
tableData = []
index = 0
for row in statsResults:
feature = row[0]
seq1 = row[1]
seq2 = row[2]
parentSeq1 = row[3]
parentSeq2 = row[4]
p1 = float(seq1) / parentSeq1
p2 = float(seq2) / parentSeq2
powerList = []
powerListLess5 = []
powerListGreater5 = []
for trial in xrange(0, trials):
if progress != '':
index += 1
progress.setValue(index)
progress.setLabelText(feature + ' - Trial = ' + str(trial))
power = 0
processedReplicates = 0
for dummy in xrange(0, bootstrapRep):
c1 = 0
c2 = 0
for dummy in xrange(0, parentSeq1):
rnd = random.random()
if rnd <= p1:
c1 += 1
for dummy in xrange(0, parentSeq2):
rnd = random.random()
if rnd <= p2:
c2 += 1
if c1 == 0 and c2 == 0:
# This is a special case that many hypothesis test will not handle correctly
# so we just ignore it. This will have little effect on the calculated power
# of a test.
continue
processedReplicates += 1
pValueOneSided, pValueTwoSided = test.hypothesisTest(c1, c2, parentSeq1, parentSeq2)
if pValueTwoSided < signLevel:
power += 1
if processedReplicates > 0:
if min([seq1,seq2]) <= 5:
powerListLess5.append(float(power) / processedReplicates)
else:
powerListGreater5.append(float(power) / processedReplicates)
powerList.append(float(power) / processedReplicates)
row = []
row.append(feature)
row.append(seq1)
row.append(seq2)
row.append(parentSeq1)
row.append(parentSeq2)
row.append(float(seq1) / parentSeq1)
row.append(float(seq2) / parentSeq2)
row.append(mean(powerList))
row.append(stdDev(powerList))
if math.isnan(mean(powerListLess5)):
row.append('')
else:
row.append(mean(powerListLess5))
if math.isnan(stdDev(powerListLess5)):
row.append('')
else:
row.append(stdDev(powerListLess5))
if math.isnan(mean(powerListGreater5)):
row.append('')
else:
row.append(mean(powerListGreater5))
if math.isnan(stdDev(powerListGreater5)):
row.append('')
else:
row.append(stdDev(powerListGreater5))
tableData.append(row)
return tableData
ciLengthRP.append(upperCI - lowerCI)
coverageListDP.append(float(containedRepDP) / replicates)
coverageListDPCC.append(float(containedRepDPCC) / replicates)
coverageListNW.append(float(containedRepNW) / replicates)
coverageListWoolf.append(float(containedRepWoolf) / replicates)
coverageListGart.append(float(containedRepGart) / replicates)
coverageListRP.append(float(containedRepRP) / replicates)
results = [coverageListDP, coverageListDPCC, coverageListNW, coverageListWoolf, coverageListGart, coverageListRP]
lengths = [ciLengthDP,ciLengthDPCC,ciLengthNW,ciLengthWoolf,ciLengthGart,ciLengthRP]
methodNames = ['DP: Asymptotic', 'DP: Asymptotic-CC', 'Newcombe-Wilson', 'Woolf', 'Gart', 'RP: Asympototic']
for i in xrange(0, len(results)):
coverageMeanStr = '%.2f' % mean(results[i])
coverageSdStr = '%.2f' % stdDev(results[i])
coverageMinStr = '%.2f' % min(results[i])
coverageMaxStr = '%.2f' % max(results[i])
lengthMeanStr = '%.2f' % mean(lengths[i])
lengthSdStr = '%.2f' % stdDev(lengths[i])
fout.write(methodNames[i] + '\n')
fout.write(coverageMeanStr + '+/-' + coverageSdStr + '[' + coverageMinStr + ';' + coverageMaxStr + ']\n')
fout.write(lengthMeanStr + '+/-' + lengthSdStr+ '\n')
fout.write('\n')
fout.close()
def run(self,
statTest,
testType,
confIntervMethod,
coverage,
profile,
progress=None):
self.results.test = statTest.name
self.results.testType = testType
self.results.alpha = 1.0 - coverage
self.results.confIntervMethod = confIntervMethod
self.results.profile = profile
if progress == 'Verbose':
print ' Processing feature:'
self.results.data = []
index = 0
# calculate statistics
seqsGroup1 = []
seqsGroup2 = []
parentSeqsGroup1 = []
parentSeqsGroup2 = []
pValues = []
lowerCIs = []
upperCIs = []
effectSizes = []
notes = []
if statTest.bSingleFeatureInterface:
# process features one at a time
for feature in profile.getFeatures():
if progress == 'Verbose':
print ' ' + feature
elif progress != None:
if progress.wasCanceled():
self.results.data = []
return
index += 1
progress.setValue(index)
# get statistics
seqGroup1, seqGroup2 = profile.getFeatureCounts(feature)
parentSeqGroup1, parentSeqGroup2 = profile.getParentFeatureCounts(
feature)
results = statTest.run(seqGroup1, seqGroup2, parentSeqGroup1,
parentSeqGroup2, confIntervMethod,
coverage)
pValueOneSided, pValueTwoSided, lowerCI, upperCI, effectSize, note = results
if testType == 'One-sided':
pValue = pValueOneSided
elif testType == 'Two-sided':
pValue = pValueTwoSided
else:
print 'Error: Unknown test type.'
# record results
seqsGroup1.append(seqGroup1)
seqsGroup2.append(seqGroup2)
parentSeqsGroup1.append(parentSeqGroup1)
parentSeqsGroup2.append(parentSeqGroup2)
pValues.append(pValue)
lowerCIs.append(lowerCI)
upperCIs.append(upperCI)
effectSizes.append(effectSize)
notes.append(note)
if progress != None and progress != 'Verbose':
index += 1
progress.setValue(index)
else:
# process all features at once
seqsGroup1, seqsGroup2 = profile.getFeatureCountsAll()
parentSeqsGroup1, parentSeqsGroup2 = profile.getParentFeatureCountsAll(
)
pValuesOneSided, pValuesTwoSided, lowerCIs, upperCIs, effectSizes, notes = statTest.runAll(
seqsGroup1, seqsGroup2, parentSeqsGroup1, parentSeqsGroup2,
confIntervMethod, coverage, progress)
if progress == 'Verbose':
print ' Processing all features...'
elif progress != None and progress.wasCanceled():
self.results.data = []
return
if testType == 'One-sided':
pValues = pValuesOneSided
elif testType == 'Two-sided':
pValues = pValuesTwoSided
else:
print 'Error: Unknown test type.'
# record statistics
features = profile.getFeatures()
for i in xrange(0, len(features)):
propGroup1 = []
for j in xrange(0, len(seqsGroup1[i])):
sg1 = seqsGroup1[i][j]
psg1 = parentSeqsGroup1[i][j]
if psg1 > 0:
propGroup1.append(sg1 * 100.0 / psg1)
else:
propGroup1.append(0.0)
propGroup2 = []
for j in xrange(0, len(seqsGroup2[i])):
sg2 = seqsGroup2[i][j]
psg2 = parentSeqsGroup2[i][j]
if psg2 > 0:
propGroup2.append(sg2 * 100.0 / psg2)
else:
propGroup2.append(0.0)
meanGroup1 = mean(propGroup1)
meanGroup2 = mean(propGroup2)
row = [
features[i], meanGroup1,
stdDev(propGroup1, meanGroup1), meanGroup2,
stdDev(propGroup2, meanGroup2),
float(pValues[i]),
float(pValues[i]),
float(effectSizes[i]),
float(lowerCIs[i]),
float(upperCIs[i]), notes[i]
]
for j in xrange(0, len(seqsGroup1[i])):
row.append(seqsGroup1[i][j])
row.append(parentSeqsGroup1[i][j])
if parentSeqsGroup1[i][j] > 0:
row.append(seqsGroup1[i][j] * 100.0 /
parentSeqsGroup1[i][j])
else:
row.append(0.0)
for j in xrange(0, len(seqsGroup2[i])):
row.append(seqsGroup2[i][j])
row.append(parentSeqsGroup2[i][j])
if parentSeqsGroup2[i][j] > 0:
row.append(seqsGroup2[i][j] * 100.0 /
parentSeqsGroup2[i][j])
else:
row.append(0.0)
self.results.data.append(row)
headingsSampleStats = []
for sampleName in (profile.samplesInGroup1 + profile.samplesInGroup2):
headingsSampleStats.append(sampleName)
headingsSampleStats.append(sampleName + ': parent seq. count')
headingsSampleStats.append(sampleName + ': rel. freq. (%)')
self.results.createTableHeadings(profile.groupName1,
profile.groupName2,
headingsSampleStats)
# sort results according to p-values
if len(self.results.data) >= 1:
self.results.data = TableHelper.SortTable(
self.results.data, [self.results.dataHeadings['pValues']])
def run(self, confIntervMethod, coverage, tables, trials, bootstrapRep, progress):
tableData = []
index = 0
for row in tables:
feature = row[0]
seq1 = row[1]
seq2 = row[2]
parentSeq1 = row[3]
parentSeq2 = row[4]
lowerCI, upperCI, obsEffectSize = confIntervMethod.run(seq1, seq2, parentSeq1, parentSeq2, coverage)
p1 = float(seq1) / parentSeq1
p2 = float(seq2) / parentSeq2
coverageList = []
coverageListLess5 = []
coverageListGreater5 = []
for trial in xrange(0, trials):
if progress != '':
index += 1
progress.setValue(index)
progress.setLabelText(feature + ' - Trial = ' + str(trial))
containedRep = 0
for dummy in xrange(0, bootstrapRep):
c1 = binomial(parentSeq1, p1)
c2 = binomial(parentSeq2, p2)
lowerCI, upperCI, effectSize = confIntervMethod.run(c1, c2, parentSeq1, parentSeq2, coverage)
if obsEffectSize >= lowerCI and obsEffectSize <= upperCI:
containedRep += 1
if min([seq1,seq2]) <= 5:
coverageListLess5.append(float(containedRep) / bootstrapRep)
else:
coverageListGreater5.append(float(containedRep) / bootstrapRep)
coverageList.append(float(containedRep) / bootstrapRep)
row = []
row.append(feature)
row.append(seq1)
row.append(seq2)
row.append(parentSeq1)
row.append(parentSeq2)
row.append(float(seq1) / parentSeq1)
row.append(float(seq2) / parentSeq2)
row.append(mean(coverageList))
row.append(stdDev(coverageList))
if math.isnan(mean(coverageListLess5)):
row.append('')
else:
row.append(mean(coverageListLess5))
if math.isnan(stdDev(coverageListLess5)):
row.append('')
else:
row.append(stdDev(coverageListLess5))
if math.isnan(mean(coverageListGreater5)):
row.append('')
else:
row.append(mean(coverageListGreater5))
if math.isnan(stdDev(coverageListGreater5)):
row.append('')
else:
row.append(stdDev(coverageListGreater5))
tableData.append(row)
return tableData
