def eig_correlate_matrices(hic_data1, hic_data2, nvect=6, normalized=False,
savefig=None, show=False, savedata=None,
remove_bad_columns=True, **kwargs):
"""
Compare the iteractions of two Hi-C matrices using their 6 first
eigenvectors, with pearson correlation
:param hic_data1: Hi-C-data object
:param hic_data2: Hi-C-data object
:param 6 nvect: number of eigenvectors to compare
:param None savefig: path to save the plot
:param False show: displays the plot
:param False normalized: use normalized data
:param True remove_bads: computes the union of bad columns between samples
and exclude them from the comparison
:param kwargs: any argument to pass to matplotlib imshow function
:returns: matrix of correlations
"""
data1 = hic_data1.get_matrix(normalized=normalized)
data2 = hic_data2.get_matrix(normalized=normalized)
## reduce matrices to remove bad columns
if remove_bad_columns:
# union of bad columns
bads = hic_data1.bads.copy()
bads.update(hic_data2.bads)
# remove them form both matrices
for bad in sorted(bads, reverse=True):
del(data1[bad])
del(data2[bad])
for i in xrange(len(data1)):
_ = data1[i].pop(bad)
_ = data2[i].pop(bad)
# get the log
size = len(data1)
data1 = nozero_log(data1, np.log2)
data2 = nozero_log(data2, np.log2)
# get the eigenvectors
ev1, evect1 = eigh(data1)
ev2, evect2 = eigh(data2)
corr = [[0 for _ in xrange(nvect)] for _ in xrange(nvect)]
# sort eigenvectors according to their eigenvalues => first is last!!
sort_perm = ev1.argsort()
ev1.sort()
evect1 = evect1[sort_perm]
sort_perm = ev2.argsort()
ev2.sort()
evect2 = evect2[sort_perm]
# calculate Pearson correlation
for i in xrange(nvect):
for j in xrange(nvect):
corr[i][j] = abs(pearsonr(evect1[:,-i-1],
evect2[:,-j-1])[0])
# plot
axe = plt.axes([0.1, 0.1, 0.6, 0.8])
cbaxes = plt.axes([0.85, 0.1, 0.03, 0.8])
if show or savefig:
im = axe.imshow(corr, interpolation="nearest",origin='lower', **kwargs)
axe.set_xlabel('Eigen Vectors exp. 1')
axe.set_ylabel('Eigen Vectors exp. 2')
axe.set_xticks(range(nvect))
axe.set_yticks(range(nvect))
axe.set_xticklabels(range(1, nvect + 2))
axe.set_yticklabels(range(1, nvect + 2))
axe.xaxis.set_tick_params(length=0, width=0)
axe.yaxis.set_tick_params(length=0, width=0)
cbar = plt.colorbar(im, cax = cbaxes )
cbar.ax.set_ylabel('Pearson correlation', rotation=90*3,
verticalalignment='bottom')
axe2 = axe.twinx()
axe2.set_yticks(range(nvect))
axe2.set_yticklabels(['%.1f' % (e) for e in ev2[-nvect:][::-1]])
axe2.set_ylabel('corresponding Eigen Values exp. 2', rotation=90*3,
verticalalignment='bottom')
axe2.set_ylim((-0.5, nvect - 0.5))
axe2.yaxis.set_tick_params(length=0, width=0)
axe3 = axe.twiny()
axe3.set_xticks(range(nvect))
axe3.set_xticklabels(['%.1f' % (e) for e in ev1[-nvect:][::-1]])
axe3.set_xlabel('corresponding Eigen Values exp. 1')
axe3.set_xlim((-0.5, nvect - 0.5))
axe3.xaxis.set_tick_params(length=0, width=0)
axe.set_ylim((-0.5, nvect - 0.5))
axe.set_xlim((-0.5, nvect - 0.5))
if savefig:
tadbit_savefig(savefig)
if show:
plt.show()
plt.close('all')
if savedata:
out = open(savedata, 'w')
out.write('# ' + '\t'.join(['Eigen Vector %s'% i
for i in xrange(nvect)]) + '\n')
for i in xrange(nvect):
out.write('\t'.join([str(corr[i][j])
for j in xrange(nvect)]) + '\n')
out.close()
if kwargs.get('get_bads', False):
return corr, bads
else:
return corr
def draw_map(data, genome_seq, cumcs, savefig, show, one=False, clim=None,
cmap='jet', decay=False, perc=10, name=None, cistrans=None,
decay_resolution=10000, normalized=False, max_diff=None):
_ = plt.figure(figsize=(15.,12.5))
if not max_diff:
max_diff = len(data)
ax1 = plt.axes([0.34, 0.08, 0.6, 0.7205])
ax2 = plt.axes([0.07, 0.65, 0.21, 0.15])
if decay:
ax3 = plt.axes([0.07, 0.42, 0.21, 0.15])
plot_distance_vs_interactions(data, genome_seq=genome_seq, axe=ax3,
resolution=decay_resolution,
max_diff=max_diff, normalized=normalized)
ax4 = plt.axes([0.34, 0.805, 0.6, 0.04], sharex=ax1)
ax5 = plt.axes([0.34, 0.845, 0.6, 0.04], sharex=ax1)
ax6 = plt.axes([0.34, 0.885, 0.6, 0.04], sharex=ax1)
try:
minoridata = np.nanmin(data)
maxoridata = np.nanmax(data)
except AttributeError:
vals = [i for d in data for i in d if not np.isnan(i)]
minoridata = np.min(vals)
maxoridata = np.max(vals)
totaloridata = np.nansum([data[i][j] for i in xrange(len(data))
for j in xrange(i, len(data[i]))]) # may not be square
data = nozero_log(data, np.log2)
vals = np.array([i for d in data for i in d])
vals = vals[np.isfinite(vals)]
mindata = np.nanmin(vals)
maxdata = np.nanmax(vals)
diff = maxdata - mindata
posI = 0.01 if not clim else (float(clim[0]) / diff) if clim[0] != None else 0.01
posF = 1.0 if not clim else (float(clim[1]) / diff) if clim[1] != None else 1.0
if cmap == 'tadbit':
cuts = perc
cdict = {'red' : [(0.0, 0.0, 0.0)],
'green': [(0.0, 0.0, 0.0)],
'blue' : [(0.0, 0.5, 0.5)]}
prev_pos = 0
median = (np.median(vals) - mindata) / diff
for prc in np.linspace(posI, median, cuts / 2, endpoint=False):
try:
pos = (np.percentile(vals, prc * 100.) - mindata) / diff
prc = ((prc - posI) / (median - posI)) + 1. / cuts
except ValueError:
pos = prc = 0
if prev_pos >= pos:
continue
cdict['red' ].append([pos, prc, prc])
cdict['green'].append([pos, prc, prc])
cdict['blue' ].append([pos, 1, 1])
prev_pos = pos
for prc in np.linspace(median + 1. / cuts, posF, cuts / 2, endpoint=False):
try:
pos = (np.percentile(vals, prc * 100.) - mindata) / diff
prc = ((prc - median) / (posF - median))
except ValueError:
pos = prc = 0
if prev_pos >= pos:
continue
cdict['red' ].append([pos, 1.0, 1.0])
cdict['green'].append([pos, 1 - prc, 1 - prc])
cdict['blue' ].append([pos, 1 - prc, 1 - prc])
prev_pos = pos
pos = (np.percentile(vals ,97.) - mindata) / diff
cdict['red' ].append([pos, 0.1, 0.1])
cdict['green'].append([pos, 0, 0])
cdict['blue' ].append([pos, 0, 0])
cdict['red' ].append([1.0, 1, 1])
cdict['green'].append([1.0, 1, 1])
cdict['blue' ].append([1.0, 0, 0])
cmap = LinearSegmentedColormap(cmap, cdict)
clim = None
else:
cmap = plt.get_cmap(cmap)
cmap.set_bad('darkgrey', 1)
ax1.imshow(data, interpolation='none',
cmap=cmap, vmin=clim[0] if clim else None, vmax=clim[1] if clim else None)
size1 = len(data)
size2 = len(data[0])
if size1 == size2:
for i in xrange(size1):
for j in xrange(i, size2):
if np.isnan(data[i][j]):
data[i][j] = 0
data[j][i] = 0
else:
for i in xrange(size1):
for j in xrange(size2):
if np.isnan(data[i][j]):
data[i][j] = 0
#data[j][i] = data[i][j]
try:
evals, evect = eigh(data)
sort_perm = evals.argsort()
evect = evect[sort_perm]
except:
evals, evect = None, None
data = [i for d in data for i in d if not np.isnan(i)]
gradient = np.linspace(np.nanmin(data),
np.nanmax(data), max(size1, size2))
gradient = np.vstack((gradient, gradient))
h = ax2.hist(data, color='darkgrey', linewidth=2,
bins=20, histtype='step', normed=True)
_ = ax2.imshow(gradient, aspect='auto', cmap=cmap,
extent=(np.nanmin(data), np.nanmax(data) , 0, max(h[0])))
if genome_seq:
for crm in genome_seq:
ax1.vlines([cumcs[crm][0]-.5, cumcs[crm][1]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='w', linestyle='-', linewidth=1, alpha=1)
ax1.hlines([cumcs[crm][1]-.5, cumcs[crm][0]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='w', linestyle='-', linewidth=1, alpha=1)
ax1.vlines([cumcs[crm][0]-.5, cumcs[crm][1]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='k', linestyle='--')
ax1.hlines([cumcs[crm][1]-.5, cumcs[crm][0]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='k', linestyle='--')
if not one:
vals = [0]
keys = ['']
for crm in genome_seq:
vals.append(cumcs[crm][0])
keys.append(crm)
vals.append(cumcs[crm][1])
ax1.set_yticks(vals)
ax1.set_yticklabels('')
ax1.set_yticks([float(vals[i]+vals[i+1])/2
for i in xrange(len(vals) - 1)], minor=True)
ax1.set_yticklabels(keys, minor=True)
for t in ax1.yaxis.get_minor_ticks():
t.tick1On = False
t.tick2On = False
# totaloridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(totaloridata)[::-1])])[::-1].strip(',')
# minoridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(minoridata)[::-1])])[::-1].strip(',')
# maxoridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(maxoridata)[::-1])])[::-1].strip(',')
plt.figtext(0.05,0.25, ''.join([
(name + '\n') if name else '',
'Number of interactions: %s\n' % str(totaloridata),
('' if np.isnan(cistrans) else
('Percentage of cis interactions: %.0f%%\n' % (cistrans*100))),
'Min interactions: %s\n' % (minoridata),
'Max interactions: %s\n' % (maxoridata)]))
ax2.set_xlim((np.nanmin(data), np.nanmax(data)))
ax2.set_ylim((0, max(h[0])))
ax1.set_xlim ((-0.5, size1 - .5))
ax1.set_ylim ((-0.5, size2 - .5))
ax2.set_xlabel('log interaction count')
# we reduce the number of dots displayed.... we just want to see the shape
subdata = np.array(list(set([float(int(d*100))/100 for d in data])))
try:
normfit = sc_norm.pdf(subdata, np.nanmean(data), np.nanstd(data))
except AttributeError:
normfit = sc_norm.pdf(subdata, np.mean(data), np.std(data))
ax2.plot(subdata, normfit, 'w.', markersize=2.5, alpha=.4)
ax2.plot(subdata, normfit, 'k.', markersize=1.5, alpha=1)
ax2.set_title('skew: %.3f, kurtosis: %.3f' % (skew(data),
kurtosis(data)))
try:
ax4.vlines(range(size1), 0, evect[:,-1], color='k')
except (TypeError, IndexError):
pass
ax4.hlines(0, 0, size2, color='red')
ax4.set_ylabel('E1')
ax4.set_yticklabels([])
try:
ax5.vlines(range(size1), 0, evect[:,-2], color='k')
except (TypeError, IndexError):
pass
ax5.hlines(0, 0, size2, color='red')
ax5.set_ylabel('E2')
ax5.set_yticklabels([])
try:
ax6.vlines(range(size1), 0, evect[:,-3], color='k')
except (TypeError, IndexError):
pass
ax6.hlines(0, 0, size2, color='red')
ax6.set_ylabel('E3')
ax6.set_yticklabels([])
xticklabels = ax4.get_xticklabels() + ax5.get_xticklabels() + ax6.get_xticklabels()
plt.setp(xticklabels, visible=False)
if savefig:
tadbit_savefig(savefig)
elif show:
plt.show()
plt.close('all')
def eig_correlate_matrices(hic_data1, hic_data2, nvect=6,
savefig=None, show=False, savedata=None, **kwargs):
"""
Compare the iteractions of two Hi-C matrices using their 6 first
eigenvectors, with spearman rank correlation
:param hic_data1: Hi-C-data object
:param hic_data2: Hi-C-data object
:param 6 nvect: number of eigenvectors to compare
:param None savefig: path to save the plot
:param False show: displays the plot
:param kwargs: any argument to pass to matplotlib imshow function
:returns: matrix of correlations
"""
data1 = hic_data1.get_matrix()
data2 = hic_data2.get_matrix()
# get the log
size = len(data1)
data1 = nozero_log(data1, np.log2)
data2 = nozero_log(data2, np.log2)
# get the eigenvectors
ev1, evect1 = eigh(data1)
ev2, evect2 = eigh(data2)
corr = [[0 for _ in xrange(nvect)] for _ in xrange(nvect)]
# sort eigenvectors according to their eigenvalues => first is last!!
sort_perm = ev1.argsort()
ev1.sort()
evect1 = evect1[sort_perm]
sort_perm = ev2.argsort()
ev2.sort()
evect2 = evect2[sort_perm]
# calculate Pearson correlation
for i in xrange(nvect):
for j in xrange(nvect):
corr[i][j] = abs(pearsonr(evect1[:,-i-1],
evect2[:,-j-1])[0])
# plot
axe = plt.axes([0.1, 0.1, 0.6, 0.8])
cbaxes = plt.axes([0.85, 0.1, 0.03, 0.8])
if show or savefig:
im = axe.imshow(corr, interpolation="nearest",origin='lower', **kwargs)
axe.set_xlabel('Eigen Vectors exp. 1')
axe.set_ylabel('Eigen Vectors exp. 2')
axe.set_xticks(range(nvect))
axe.set_yticks(range(nvect))
axe.set_xticklabels(range(1, nvect + 2))
axe.set_yticklabels(range(1, nvect + 2))
axe.xaxis.set_tick_params(length=0, width=0)
axe.yaxis.set_tick_params(length=0, width=0)
cbar = plt.colorbar(im, cax = cbaxes )
cbar.ax.set_ylabel('Pearson correlation', rotation=90*3,
verticalalignment='bottom')
axe2 = axe.twinx()
axe2.set_yticks(range(nvect))
axe2.set_yticklabels(['%.1f' % (e) for e in ev2[-nvect:][::-1]])
axe2.set_ylabel('corresponding Eigen Values exp. 2', rotation=90*3,
verticalalignment='bottom')
axe2.set_ylim((-0.5, nvect - 0.5))
axe2.yaxis.set_tick_params(length=0, width=0)
axe3 = axe.twiny()
axe3.set_xticks(range(nvect))
axe3.set_xticklabels(['%.1f' % (e) for e in ev1[-nvect:][::-1]])
axe3.set_xlabel('corresponding Eigen Values exp. 1')
axe3.set_xlim((-0.5, nvect - 0.5))
axe3.xaxis.set_tick_params(length=0, width=0)
axe.set_ylim((-0.5, nvect - 0.5))
axe.set_xlim((-0.5, nvect - 0.5))
if savefig:
tadbit_savefig(savefig)
if show:
plt.show()
plt.close('all')
if savedata:
out = open(savedata, 'w')
out.write('# ' + '\t'.join(['Eigen Vector %s'% i
for i in xrange(nvect)]) + '\n')
for i in xrange(nvect):
out.write('\t'.join([str(corr[i][j])
for j in xrange(nvect)]) + '\n')
out.close()
return corr
def draw_map(data, genome_seq, cumcs, savefig, show, one=False, clim=None,
cmap='jet', decay=False, perc=10, name=None, cistrans=None,
decay_resolution=10000, normalized=False, max_diff=None):
_ = plt.figure(figsize=(15.,12.5))
if not max_diff:
max_diff = len(data)
ax1 = plt.axes([0.34, 0.08, 0.6, 0.7205])
ax2 = plt.axes([0.07, 0.65, 0.21, 0.15])
if decay:
ax3 = plt.axes([0.07, 0.42, 0.21, 0.15])
plot_distance_vs_interactions(data, genome_seq=genome_seq, axe=ax3,
resolution=decay_resolution,
max_diff=max_diff, normalized=normalized)
ax4 = plt.axes([0.34, 0.805, 0.6, 0.04], sharex=ax1)
ax5 = plt.axes([0.34, 0.845, 0.6, 0.04], sharex=ax1)
ax6 = plt.axes([0.34, 0.885, 0.6, 0.04], sharex=ax1)
try:
minoridata = np.nanmin(data)
maxoridata = np.nanmax(data)
except AttributeError:
vals = [i for d in data for i in d if not np.isnan(i)]
minoridata = np.min(vals)
maxoridata = np.max(vals)
totaloridata = np.nansum([data[i][j] for i in xrange(len(data))
for j in xrange(i, len(data))])
data = nozero_log(data, np.log2)
vals = np.array([i for d in data for i in d])
vals = vals[np.isfinite(vals)]
mindata = np.nanmin(vals)
maxdata = np.nanmax(vals)
diff = maxdata - mindata
posI = 0.01 if not clim else (float(clim[0]) / diff) if clim[0] != None else 0.01
posF = 1.0 if not clim else (float(clim[1]) / diff) if clim[1] != None else 1.0
if cmap == 'tadbit':
cuts = perc
cdict = {'red' : [(0.0, 0.0, 0.0)],
'green': [(0.0, 0.0, 0.0)],
'blue' : [(0.0, 0.5, 0.5)]}
prev_pos = 0
median = (np.median(vals) - mindata) / diff
for prc in np.linspace(posI, median, cuts / 2, endpoint=False):
try:
pos = (np.percentile(vals, prc * 100.) - mindata) / diff
prc = ((prc - posI) / (median - posI)) + 1. / cuts
except ValueError:
pos = prc = 0
if prev_pos >= pos:
continue
cdict['red' ].append([pos, prc, prc])
cdict['green'].append([pos, prc, prc])
cdict['blue' ].append([pos, 1, 1])
prev_pos = pos
for prc in np.linspace(median + 1. / cuts, posF, cuts / 2, endpoint=False):
try:
pos = (np.percentile(vals, prc * 100.) - mindata) / diff
prc = ((prc - median) / (posF - median))
except ValueError:
pos = prc = 0
if prev_pos >= pos:
continue
cdict['red' ].append([pos, 1.0, 1.0])
cdict['green'].append([pos, 1 - prc, 1 - prc])
cdict['blue' ].append([pos, 1 - prc, 1 - prc])
prev_pos = pos
pos = (np.percentile(vals ,97.) - mindata) / diff
cdict['red' ].append([pos, 0.1, 0.1])
cdict['green'].append([pos, 0, 0])
cdict['blue' ].append([pos, 0, 0])
cdict['red' ].append([1.0, 1, 1])
cdict['green'].append([1.0, 1, 1])
cdict['blue' ].append([1.0, 0, 0])
cmap = LinearSegmentedColormap(cmap, cdict)
clim = None
else:
cmap = plt.get_cmap(cmap)
cmap.set_bad('darkgrey', 1)
ax1.imshow(data, interpolation='none',
cmap=cmap, vmin=clim[0] if clim else None, vmax=clim[1] if clim else None)
size = len(data)
for i in xrange(size):
for j in xrange(i, size):
if np.isnan(data[i][j]):
data[i][j] = 0
data[j][i] = 0
#data[j][i] = data[i][j]
evals, evect = eigh(data)
sort_perm = evals.argsort()
evect = evect[sort_perm]
data = [i for d in data for i in d if not np.isnan(i)]
gradient = np.linspace(np.nanmin(data),
np.nanmax(data), size)
gradient = np.vstack((gradient, gradient))
h = ax2.hist(data, color='darkgrey', linewidth=2,
bins=20, histtype='step', normed=True)
_ = ax2.imshow(gradient, aspect='auto', cmap=cmap,
extent=(np.nanmin(data), np.nanmax(data) , 0, max(h[0])))
if genome_seq:
for crm in genome_seq:
ax1.vlines([cumcs[crm][0]-.5, cumcs[crm][1]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='w', linestyle='-', linewidth=1, alpha=1)
ax1.hlines([cumcs[crm][1]-.5, cumcs[crm][0]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='w', linestyle='-', linewidth=1, alpha=1)
ax1.vlines([cumcs[crm][0]-.5, cumcs[crm][1]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='k', linestyle='--')
ax1.hlines([cumcs[crm][1]-.5, cumcs[crm][0]-.5], cumcs[crm][0]-.5, cumcs[crm][1]-.5,
color='k', linestyle='--')
if not one:
vals = [0]
keys = ['']
for crm in genome_seq:
vals.append(cumcs[crm][0])
keys.append(crm)
vals.append(cumcs[crm][1])
ax1.set_yticks(vals)
ax1.set_yticklabels('')
ax1.set_yticks([float(vals[i]+vals[i+1])/2
for i in xrange(len(vals) - 1)], minor=True)
ax1.set_yticklabels(keys, minor=True)
for t in ax1.yaxis.get_minor_ticks():
t.tick1On = False
t.tick2On = False
# totaloridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(totaloridata)[::-1])])[::-1].strip(',')
# minoridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(minoridata)[::-1])])[::-1].strip(',')
# maxoridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(maxoridata)[::-1])])[::-1].strip(',')
plt.figtext(0.05,0.25, ''.join([
(name + '\n') if name else '',
'Number of interactions: %s\n' % str(totaloridata),
('' if np.isnan(cistrans) else
('Percentage of cis interactions: %.0f%%\n' % (cistrans*100))),
'Min interactions: %s\n' % (minoridata),
'Max interactions: %s\n' % (maxoridata)]))
ax2.set_xlim((np.nanmin(data), np.nanmax(data)))
ax2.set_ylim((0, max(h[0])))
ax1.set_xlim ((-0.5, size - .5))
ax1.set_ylim ((-0.5, size - .5))
ax2.set_xlabel('log interaction count')
# we reduce the number of dots displayed.... we just want to see the shape
subdata = np.array(list(set([float(int(d*100))/100 for d in data])))
try:
normfit = sc_norm.pdf(subdata, np.nanmean(data), np.nanstd(data))
except AttributeError:
normfit = sc_norm.pdf(subdata, np.mean(data), np.std(data))
ax2.plot(subdata, normfit, 'w.', markersize=2.5, alpha=.4)
ax2.plot(subdata, normfit, 'k.', markersize=1.5, alpha=1)
ax2.set_title('skew: %.3f, kurtosis: %.3f' % (skew(data),
kurtosis(data)))
ax4.vlines(range(size), 0, evect[:,-1], color='k')
ax4.hlines(0, 0, size, color='red')
ax4.set_ylabel('E1')
ax4.set_yticklabels([])
try:
ax5.vlines(range(size), 0, evect[:,-2], color='k')
except IndexError:
pass
ax5.hlines(0, 0, size, color='red')
ax5.set_ylabel('E2')
ax5.set_yticklabels([])
try:
ax6.vlines(range(size), 0, evect[:,-3], color='k')
except IndexError:
pass
ax6.hlines(0, 0, size, color='red')
ax6.set_ylabel('E3')
ax6.set_yticklabels([])
xticklabels = ax4.get_xticklabels() + ax5.get_xticklabels() + ax6.get_xticklabels()
plt.setp(xticklabels, visible=False)
if savefig:
tadbit_savefig(savefig)
elif show:
plt.show()
plt.close('all')
def eig_correlate_matrices(hic_data1, hic_data2, nvect=6,
savefig=None, show=False, savedata=None, **kwargs):
"""
Compare the iteractions of two Hi-C matrices using their 6 first
eigenvectors, with spearman rank correlation
:param hic_data1: Hi-C-data object
:param hic_data2: Hi-C-data object
:param 6 nvect: number of eigenvectors to compare
:param None savefig: path to save the plot
:param False show: displays the plot
:param kwargs: any argument to pass to matplotlib imshow function
:returns: matrix of correlations
"""
data1 = hic_data1.get_matrix()
data2 = hic_data2.get_matrix()
# get the log
size = len(data1)
data1 = nozero_log(data1, np.log2)
data2 = nozero_log(data2, np.log2)
# get the eigenvectors
ev1, evect1 = eigh(data1)
ev2, evect2 = eigh(data2)
corr = [[0 for _ in xrange(nvect)] for _ in xrange(nvect)]
# sort eigenvectors according to their eigenvalues => first is last!!
sort_perm = ev1.argsort()
ev1.sort()
evect1 = evect1[sort_perm]
sort_perm = ev2.argsort()
ev2.sort()
evect2 = evect2[sort_perm]
# calculate Pearson correlation
for i in xrange(nvect):
for j in xrange(nvect):
corr[i][j] = abs(pearsonr(evect1[:,-i-1],
evect2[:,-j-1])[0])
# plot
axe = plt.axes([0.1, 0.1, 0.6, 0.8])
cbaxes = plt.axes([0.85, 0.1, 0.03, 0.8])
if show or savefig:
im = axe.imshow(corr, interpolation="nearest",origin='lower', **kwargs)
axe.set_xlabel('Eigen Vectors exp. 1')
axe.set_ylabel('Eigen Vectors exp. 2')
axe.set_xticks(range(nvect))
axe.set_yticks(range(nvect))
axe.set_xticklabels(range(1, nvect + 2))
axe.set_yticklabels(range(1, nvect + 2))
axe.xaxis.set_tick_params(length=0, width=0)
axe.yaxis.set_tick_params(length=0, width=0)
cbar = plt.colorbar(im, cax = cbaxes )
cbar.ax.set_ylabel('Pearson correlation', rotation=90*3,
verticalalignment='bottom')
axe2 = axe.twinx()
axe2.set_yticks(range(nvect))
axe2.set_yticklabels(['%.1f' % (e) for e in ev2[-nvect:][::-1]])
axe2.set_ylabel('corresponding Eigen Values exp. 2', rotation=90*3,
verticalalignment='bottom')
axe2.set_ylim((-0.5, nvect - 0.5))
axe2.yaxis.set_tick_params(length=0, width=0)
axe3 = axe.twiny()
axe3.set_xticks(range(nvect))
axe3.set_xticklabels(['%.1f' % (e) for e in ev1[-nvect:][::-1]])
axe3.set_xlabel('corresponding Eigen Values exp. 1')
axe3.set_xlim((-0.5, nvect - 0.5))
axe3.xaxis.set_tick_params(length=0, width=0)
axe.set_ylim((-0.5, nvect - 0.5))
axe.set_xlim((-0.5, nvect - 0.5))
if savefig:
tadbit_savefig(savefig)
if show:
plt.show()
plt.close('all')
if savedata:
out = open(savedata, 'w')
out.write('# ' + '\t'.join(['Eigen Vector %s'% i
for i in xrange(nvect)]) + '\n')
for i in xrange(nvect):
out.write('\t'.join([str(corr[i][j])
for j in xrange(nvect)]) + '\n')
out.close()
return corr
def draw_map(
data,
genome_seq,
cumcs,
savefig,
show,
one=False,
clim=None,
cmap="jet",
decay=False,
perc=10,
name=None,
cistrans=None,
decay_resolution=10000,
normalized=False,
max_diff=None,
):
_ = plt.figure(figsize=(15.0, 12.5))
if not max_diff:
max_diff = len(data)
ax1 = plt.axes([0.34, 0.08, 0.6, 0.7205])
ax2 = plt.axes([0.07, 0.65, 0.21, 0.15])
if decay:
ax3 = plt.axes([0.07, 0.42, 0.21, 0.15])
plot_distance_vs_interactions(
data, genome_seq=genome_seq, axe=ax3, resolution=decay_resolution, max_diff=max_diff, normalized=normalized
)
ax4 = plt.axes([0.34, 0.805, 0.6, 0.04], sharex=ax1)
ax5 = plt.axes([0.34, 0.845, 0.6, 0.04], sharex=ax1)
ax6 = plt.axes([0.34, 0.885, 0.6, 0.04], sharex=ax1)
try:
minoridata = np.nanmin(data)
maxoridata = np.nanmax(data)
except AttributeError:
vals = [i for d in data for i in d if not np.isnan(i)]
minoridata = np.min(vals)
maxoridata = np.max(vals)
totaloridata = np.nansum([data[i][j] for i in xrange(len(data)) for j in xrange(i, len(data))])
data = nozero_log(data, np.log2)
vals = np.array([i for d in data for i in d])
vals = vals[np.isfinite(vals)]
mindata = np.nanmin(vals)
maxdata = np.nanmax(vals)
diff = maxdata - mindata
posI = 0.01 if not clim else (float(clim[0]) / diff) if clim[0] != None else 0.01
posF = 1.0 if not clim else (float(clim[1]) / diff) if clim[1] != None else 1.0
if cmap == "tadbit":
cuts = perc
cdict = {"red": [(0.0, 0.0, 0.0)], "green": [(0.0, 0.0, 0.0)], "blue": [(0.0, 0.5, 0.5)]}
prev_pos = 0
median = (np.median(vals) - mindata) / diff
for prc in np.linspace(posI, median, cuts / 2, endpoint=False):
try:
pos = (np.percentile(vals, prc * 100.0) - mindata) / diff
prc = ((prc - posI) / (median - posI)) + 1.0 / cuts
except ValueError:
pos = prc = 0
if prev_pos >= pos:
continue
cdict["red"].append([pos, prc, prc])
cdict["green"].append([pos, prc, prc])
cdict["blue"].append([pos, 1, 1])
prev_pos = pos
for prc in np.linspace(median + 1.0 / cuts, posF, cuts / 2, endpoint=False):
try:
pos = (np.percentile(vals, prc * 100.0) - mindata) / diff
prc = (prc - median) / (posF - median)
except ValueError:
pos = prc = 0
if prev_pos >= pos:
continue
cdict["red"].append([pos, 1.0, 1.0])
cdict["green"].append([pos, 1 - prc, 1 - prc])
cdict["blue"].append([pos, 1 - prc, 1 - prc])
prev_pos = pos
pos = (np.percentile(vals, 97.0) - mindata) / diff
cdict["red"].append([pos, 0.1, 0.1])
cdict["green"].append([pos, 0, 0])
cdict["blue"].append([pos, 0, 0])
cdict["red"].append([1.0, 1, 1])
cdict["green"].append([1.0, 1, 1])
cdict["blue"].append([1.0, 0, 0])
cmap = LinearSegmentedColormap(cmap, cdict)
clim = None
else:
cmap = plt.get_cmap(cmap)
cmap.set_bad("darkgrey", 1)
ax1.imshow(data, interpolation="none", cmap=cmap, vmin=clim[0] if clim else None, vmax=clim[1] if clim else None)
size = len(data)
for i in xrange(size):
for j in xrange(i, size):
if np.isnan(data[i][j]):
data[i][j] = 0
data[j][i] = 0
# data[j][i] = data[i][j]
evals, evect = eigh(data)
sort_perm = evals.argsort()
evect = evect[sort_perm]
data = [i for d in data for i in d if not np.isnan(i)]
gradient = np.linspace(np.nanmin(data), np.nanmax(data), size)
gradient = np.vstack((gradient, gradient))
h = ax2.hist(data, color="darkgrey", linewidth=2, bins=20, histtype="step", normed=True)
_ = ax2.imshow(gradient, aspect="auto", cmap=cmap, extent=(np.nanmin(data), np.nanmax(data), 0, max(h[0])))
if genome_seq:
for crm in genome_seq:
ax1.vlines(
[cumcs[crm][0] - 0.5, cumcs[crm][1] - 0.5],
cumcs[crm][0] - 0.5,
cumcs[crm][1] - 0.5,
color="w",
linestyle="-",
linewidth=1,
alpha=1,
)
ax1.hlines(
[cumcs[crm][1] - 0.5, cumcs[crm][0] - 0.5],
cumcs[crm][0] - 0.5,
cumcs[crm][1] - 0.5,
color="w",
linestyle="-",
linewidth=1,
alpha=1,
)
ax1.vlines(
[cumcs[crm][0] - 0.5, cumcs[crm][1] - 0.5],
cumcs[crm][0] - 0.5,
cumcs[crm][1] - 0.5,
color="k",
linestyle="--",
)
ax1.hlines(
[cumcs[crm][1] - 0.5, cumcs[crm][0] - 0.5],
cumcs[crm][0] - 0.5,
cumcs[crm][1] - 0.5,
color="k",
linestyle="--",
)
if not one:
vals = [0]
keys = [""]
for crm in genome_seq:
vals.append(cumcs[crm][0])
keys.append(crm)
vals.append(cumcs[crm][1])
ax1.set_yticks(vals)
ax1.set_yticklabels("")
ax1.set_yticks([float(vals[i] + vals[i + 1]) / 2 for i in xrange(len(vals) - 1)], minor=True)
ax1.set_yticklabels(keys, minor=True)
for t in ax1.yaxis.get_minor_ticks():
t.tick1On = False
t.tick2On = False
# totaloridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(totaloridata)[::-1])])[::-1].strip(',')
# minoridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(minoridata)[::-1])])[::-1].strip(',')
# maxoridata = ''.join([j + ('' if (i+1)%3 else ',') for i, j in enumerate(str(maxoridata)[::-1])])[::-1].strip(',')
plt.figtext(
0.05,
0.25,
"".join(
[
(name + "\n") if name else "",
"Number of interactions: %s\n" % str(totaloridata),
("" if np.isnan(cistrans) else ("Percentage of cis interactions: %.0f%%\n" % (cistrans * 100))),
"Min interactions: %s\n" % (minoridata),
"Max interactions: %s\n" % (maxoridata),
]
),
)
ax2.set_xlim((np.nanmin(data), np.nanmax(data)))
ax2.set_ylim((0, max(h[0])))
ax1.set_xlim((-0.5, size - 0.5))
ax1.set_ylim((-0.5, size - 0.5))
ax2.set_xlabel("log interaction count")
# we reduce the number of dots displayed.... we just want to see the shape
subdata = np.array(list(set([float(int(d * 100)) / 100 for d in data])))
try:
normfit = sc_norm.pdf(subdata, np.nanmean(data), np.nanstd(data))
except AttributeError:
normfit = sc_norm.pdf(subdata, np.mean(data), np.std(data))
ax2.plot(subdata, normfit, "w.", markersize=2.5, alpha=0.4)
ax2.plot(subdata, normfit, "k.", markersize=1.5, alpha=1)
ax2.set_title("skew: %.3f, kurtosis: %.3f" % (skew(data), kurtosis(data)))
ax4.vlines(range(size), 0, evect[:, -1], color="k")
ax4.hlines(0, 0, size, color="red")
ax4.set_ylabel("E1")
ax4.set_yticklabels([])
try:
ax5.vlines(range(size), 0, evect[:, -2], color="k")
except IndexError:
pass
ax5.hlines(0, 0, size, color="red")
ax5.set_ylabel("E2")
ax5.set_yticklabels([])
try:
ax6.vlines(range(size), 0, evect[:, -3], color="k")
except IndexError:
pass
ax6.hlines(0, 0, size, color="red")
ax6.set_ylabel("E3")
ax6.set_yticklabels([])
xticklabels = ax4.get_xticklabels() + ax5.get_xticklabels() + ax6.get_xticklabels()
plt.setp(xticklabels, visible=False)
if savefig:
tadbit_savefig(savefig)
elif show:
plt.show()
plt.close("all")