def gdraw0(graphs, plotname = 'default_name', measure = 'cosine'):
pos = nx.graphviz_layout(graphs['kg'])
adjs = [ array(nx.adj_matrix(g)) for g in graphs.values() ]
nrms = []
for a in adjs:
n = sqrt(sum(a**2))
nrms.append(a / n)
kgelt = graphs.keys().index('kg')
if measure == 'cosine':
sims = array([round(nfu.cosine_adj(a1,nrms[kgelt]),8) for a1 in nrms])
else:
raise Exception()
kg = graphs['kg']
srto = argsort(graphs.keys())
#XVALs give ranks of each key index.
xvals = argsort(srto)
cols = map(lambda x:
('flt' in x and x.count('thr') > 1) and 'orange' or
('flt' in x) and 'red' or
('thr' in x) and 'yellow' or
('fg' in x) and 'green' or
('su' in x) and 'blue' or
'black', graphs.keys())
yvals = sims
f = plt.gcf()
f = myplots.fignum(3, (.25 * len(sims),10))
f.clear()
ax = f.add_subplot(111)
myplots.padded_limits(ax,xvals,yvals + [0.], margin = [.02,.02])
ax.scatter(xvals,yvals,100, color = cols)
ax.set_ylabel('red fly similarity ({0})'.format(measure))
ax.set_xlabel('networks')
ax.set_xticklabels([])
ax.set_xticks([])
mark_ys = [0, median(sims), mean(sims), sort(sims)[::-1][1],1]
ax.hlines(mark_ys, *ax.get_xlim(), linestyle = ':',alpha = .2)
f.savefig(cfg.dataPath('figs/meAWG/filter_{0}_meth_{1}_nolabels.pdf'.\
format(plotname,measure)))
ax.set_xticks(range(len(srto)))
ax.annotate('\n'.join(' '.join(z) for z in zip(graphs.keys(),cols)),
[0,1],xycoords = 'axes fraction', va = 'top')
ax.set_xticklabels([graphs.keys()[i] for i in srto],
rotation = 45, size = 'xx-small',ha = 'right')
f.savefig(cfg.dataPath('figs/meAWG/filter_{0}_meth_{1}_labels.pdf'.\
format(plotname,measure)))
def gdraw(bgraph,cgraphs, plotname = 'default_name', measure = 'cosine'):
#pos = nx.graphviz_layout(cgraphs['kg'])
nodelist = bgraph.nodes()
adjs = [ array(nx.adj_matrix(g, nodelist = nodelist)) for g in cgraphs.values() ]
badj = array(nx.adj_matrix(bgraph, nodelist = nodelist))
bnrm = badj / sqrt(sum(badj**2))
if measure == 'cosine':
nrms = []
bnrm = badj / sqrt(sum(badj**2))
for a in adjs:
n = sqrt(sum(a**2))
nrms.append(a / n)
sims = array([round(nfu.cosine_adj(a1,bnrm),8) for a1 in nrms])
elif measure =='jaccard':
sims = array([round(nfu.dotprod(a1,badj),8)/ (sum(a1) + sum(badj))
for a1 in adjs])
elif measure =='specificity':
sims = array([round(nfu.dotprod(a1,badj),8)/sum(a1) for a1 in adjs])
elif measure =='sensitivity':
sims = array([round(nfu.dotprod(a1,badj),8)/sum(badj) for a1 in adjs])
else:
raise Exception()
srto = argsort(cgraphs.keys())
#XVALs give ranks of each key index.
xvals = argsort(srto)
cols = map(lambda x:
('flt' in x and x.count('thr') > 1) and 'orange' or
('flt' in x) and 'red' or
('thr' in x) and 'yellow' or
('fg' in x) and 'green' or
('su' in x) and 'blue' or
'black', cgraphs.keys())
yvals = sims
f = plt.gcf()
f = myplots.fignum(3, (.25 * len(sims),10))
f.clear()
ax = f.add_subplot(111)
myplots.padded_limits(ax,xvals,yvals + [0.], margin = [.02,.02])
ax.scatter(xvals,yvals,100, color = cols)
ax.set_ylabel('red fly similarity ({0})'.format(measure))
ax.set_xlabel('networks')
ax.set_xticklabels([])
ax.set_xticks([])
mark_ys = [0, median(sims), mean(sims), sort(sims)[::-1][1],1]
ax.hlines(mark_ys, *ax.get_xlim(), linestyle = ':',alpha = .2)
f.savefig(cfg.dataPath('figs/meAWG/filter_{0}_meth_{1}_nolabels.pdf'.\
format(plotname,measure)))
ax.set_xticks(range(len(srto)))
cols_added = []
annotes = []
for z in zip(cgraphs.keys(),cols):
if not z[1] in cols_added:
annotes.append( ' '.join(z))
cols_added.append(z[1])
ax.annotate('\n'.join(annotes),
[1,1],xycoords = 'axes fraction', va = 'top', ha = 'right')
ax.set_xticklabels([cgraphs.keys()[i] for i in srto],
rotation = 90, va = 'bottom', size = 'xx-small')
f.savefig(cfg.dataPath('figs/meAWG/filter_{0}_meth_{1}_labels.pdf'.\
format(plotname,measure)))
def show_clusters(mods, genes, tfs,
switch_axes = False,
axes = 'space'):
mod_srt = sorted(mods.iteritems(), key =lambda x: len(x[1]))[::-1]
mod_ofinterest = mod_srt[0]
f = myplots.fignum(3, (14,8))
ids,tis, cs = [[] for i in range(3)]
n_tis = 60; ct = mycolors.getct(n_tis)
for i,t in enumerate(tsrt[:n_tis]):
nucs = [x[0] for x in t[1]['cts'] if x[1] == time_val][::5]
ids.extend(nucs)
tis.extend([i] * len(nucs))
cs.extend([ct[i] for j in range( len(nucs))])
#GET COORDS
coords = array(list(get_coords(axes = axes,
time_val = time_val,
spatial_idxs = spatial_idxs,
rows = rows,
ids = ids)))
print shape(coords)
print np.min(coords), np.max(coords)
#GET ELPSES
all_elps = cluster_elps(coords, tis)
module_scores = {}
mods_of_interest = mod_srt[:3]
for m in mods_of_interest:
mcounts = []
for i, e in enumerate(all_elps[0]):
t = tsrt[i]
tkey = t[0]
mcounts.append( len([e
for elt in m[1]
if elt['tissue'] == tkey]))
mvals = array(mcounts, float) / max(mcounts)
module_scores[m[0]] = mvals
#PLOT EM
all_axes = [f.add_subplot(ax_str) for ax_str in ['221', '222', '223', '224']]
ax0 = all_axes[0]
ax1 = all_axes[1]
cnv = mpl.colors.ColorConverter()
for j, ax in enumerate([ ax0, ax1]):
if switch_axes: myplots.padded_limits(ax, coords[0,:], coords[j+1,:])
else: myplots.padded_limits(ax, coords[0,:], coords[1,:])
elps = all_elps[j]
for i, e in enumerate(elps):
yellow =squeeze(\
mpl.colors.rgb_to_hsv( array(cnv.to_rgb('yellow'))[na, na, :]))
red = array([0.,1,1])
mscore = module_scores.values()[j][i]
hsv = yellow * (1 - mscore) + red * mscore
e.set_alpha(.75)
e.width = e.width/3
e.height = e.width/3
e.set_zorder(mscore)
color = squeeze([1,0,0])
alpha = mscore
e.set_facecolor(color)
e.set_alpha(alpha)
e.set_edgecolor('none')
ax.add_patch(e)
for annote_axis , plane in zip(*[[ax0, ax1],['XY','XZ']]):
annote_axis.set_title('\n'.join(tw.wrap('''PCA projection in gene space of'''+\
' blastoderm nuclei for time = {0}.'.format(time_val)+\
'Colors represent clusters used in'+\
'model building\n{1} Axes, {2} Plane'.\
format(time_val,axes, plane),50)))
annote_axis.set_xticks([])
annote_axis.set_yticks([])
f.savefig(myplots.figpath('first_filtering_time{0}_Axis={1}.pdf'.\
format(time_val,axes), delete = True))
raise Exception()
def show_mcmc(graphs,nc_base =12,
weighted = True, module_type = 'doubles',
measure = 'cosine'):
nodelist = graphs.values()[0].nodes()
print 'using module type: {0}'.format(module_type)
adjs = [ array(nx.adj_matrix(g, nodelist = nodelist)) for g in graphs.values() ]
nrms = []
for a in adjs:
n = sqrt(sum(a**2))
nrms.append(a / n)
idxs_allowed =[ i for i, k in enumerate(graphs.keys()) if 'mcmc' in k]
belt = graphs.keys().index('network_flt{0}'.format(nc_base))
if measure == 'cosine':
sims = array([round(nfu.cosine_adj(a1,nrms[belt]),8)
for i, a1 in enumerate(nrms) if i in idxs_allowed])
elif measure == 'dotprod':
sims = array([nfu.dotprod(a1,adjs[belt])
for i, a1 in enumerate(adfs) if i in idxs_allowed])
else:
raise Exception()
keys_allowed = [k for i, k in enumerate(graphs.keys())
if i in idxs_allowed]
srto = argsort([k for i, k in enumerate(graphs.keys())
if i in idxs_allowed])
#XVALs give ranks of each key index.
xvals = argsort(srto)
cols = map(lambda x:
('flt' in x and x.count('thr') > 1) and 'orange' or
('flt' in x) and 'red' or
('thr' in x) and 'yellow' or
('fg' in x) and 'green' or
('su' in x) and 'blue' or
'black', keys_allowed)
yvals = sims
f = plt.gcf()
f = myplots.fignum(3, (.25 * len(sims),10))
f.clear()
ax = f.add_subplot(111)
myplots.padded_limits(ax,xvals,yvals + [0.], margin = [.02,.02])
ax.scatter(xvals,yvals,100, color = cols)
ax.set_ylabel('red fly similarity ({0})'.format(measure))
ax.set_xlabel('networks')
ax.set_xticklabels([])
ax.set_xticks([])
mark_ys = [0, median(sims), mean(sims), sort(sims)[::-1][1],1]
ax.hlines(mark_ys, *ax.get_xlim(), linestyle = ':',alpha = .2)
ax.vlines(range(len(xvals))[::10], *ax.get_ylim(), linestyle = ':',alpha = .1)
figtitle = 'mcmc_net_comparisons_{0}_{1}net_comps{2}_sim_{3}_nolabels'.\
format(module_type,nc_base ,'' if weighted else 'unweighted',measure)
f.savefig(figtemplate.format(figtitle))
ax.set_xticks(range(len(srto)))
#ax.annotate('\n'.join(' '.join(z) for z in zip(graphs.keys())),
# [0,1],xycoords = 'axes fraction', va = 'top')
ax.set_xticklabels([keys_allowed[i] for i in srto],
rotation = 90, size = 'xx-small',
)#color = cols)
figtitle = 'mcmc_net_comparisons_{0}_{1}net_comps{2}_sim_{3}_labels'.\
format(module_type,nc_base ,'' if weighted else 'unweighted', measure)
f.savefig(figtemplate.format(figtitle))
def view0(modules,
data_src = 'bdtnp',
net_src = 'fRN',
max_rank = 4,
module_type = 'doubles'):
'''
A routine to view the sign of interaction coefficients for
a given transcription factor split per-cluster and per-module
size.
Designed to be run on the output of view_output.modules()
'''
#COMPUTE BULK STATISTICS FOR EACH TF
bd_data = nio.getBDTNP()
genes = bd_data.keys()
tfs =sorted(set(it.chain(*[k for k in modules.keys()])))
tf_net = nx.Graph()
tf_net.add_nodes_from(tfs)
tf_edges = it.chain(*[[(e0, e1)
for e0 in term
for e1 in term
if e0 != e1]
for term in modules.keys()])
tf_net.add_edges_from(tf_edges)
pos = nx.graphviz_layout(tf_net)
fig = myplots.fignum(1, (8,8))
tfnodes = tf_net.nodes()
tfnames = tfnodes
all_coefs = \
list(it.chain(*[t['coefs'] for t in modules.values()]))
cstd = std(all_coefs)
def colorfun(coefs):
coef = median(coefs)
arr = array([1.,0.,0.]) if coef < 0\
else array([0.,0.,1.])
return arr*min([1, abs(coef/cstd*2)])
def widthfun(coefs, maxwid):
return max([1,5.* len(coefs) /maxwid])
for tf in tfs:
fig.clf()
ax = fig.add_subplot(111)
ecols,ewids, estyles, ealphas =\
[{} for i in range(4)]
edges = []
tf_doublet_terms = [(k, v)
for k, v in modules.iteritems()
if tf in k and len(set(k)) == 2 ]
tf_triplet_terms = [(k, v)
for k, v in modules.iteritems()
if tf in k and len(set(k)) == 3 ]
isdub = dict([(k,0.) for k in tfnames])
istrip =dict([(k,0.) for k in tfnames])
coflens = [len(e[1]['coefs'])
for e in tf_triplet_terms + tf_doublet_terms]
max_l = max(coflens)
for tt,v in sorted(tf_triplet_terms,
key = lambda x: len(x[1]['genes']))[::-1][:max_rank]:
partners =tuple( [t for t in set(tt) if not t==tf])
for p in partners: istrip[p] = 1 #istrip[[tfnodes.index(p) for p in partners]] = 1
edge = partners
ecols[edge] = colorfun(modules[tt]['coefs'])
ewids[edge] = widthfun(modules[tt]['coefs'],max_l)
ealphas[edge] = 1 if module_type in ['triples','all'] \
else .1
estyles[edge] = 'dotted'
edges.append(edge)
for td,v in sorted(tf_doublet_terms,
key = lambda x: len(x[1]['genes']))[::-1][:max_rank]:
partners = tuple([t for t in set(td) if not t==tf])
for p in partners: isdub[p] = 1
edge = (tuple([tf] + list(partners)))
ecols[edge] = colorfun(modules[td]['coefs'])
ewids[edge] = widthfun(modules[td]['coefs'], max_l)
ealphas[edge] = 1 if module_type in ['doubles','all'] \
else .1
estyles[edge] = 'solid'
edges.append(edge)
tf_graph = nx.DiGraph()
tf_graph.add_nodes_from(tfnodes)
tf_graph.add_edges_from(edges)
ckw = dict([ (k, dict(color = ecols[k], #array([1,0,0])*isdub[k] +\
# array([0,0,1])*istrip[k],
alpha = ealphas[k],
linestyle = estyles[k],
linewidth = ewids[k],
arrowstyle = '-'))
for k in tf_graph.edges()])
circlepoints = dict([ (k, dict(facecolor ='white', #array([1,0,0])*isdub[k] +\
# array([0,0,1])*istrip[k],
alpha = round(ealphas[k],2),
edgecolor = ecols[k],
linestyle = estyles[k],
linewidth = 3,))
for k in tf_graph.edges()])
ax.set_title('Top modules for TF: {0}'.format(tf))
myplots.padded_limits(ax, *zip(*pos.values()))
nodes = tf_graph.nodes()
gd.draw(tf_graph,pos,edges,
scatter_nodes =tf_graph.nodes(),
skw = {'s':[200 if n == tf else 2
for n in nodes],
'facecolor':[colorfun(modules[tuple([n])]['coefs'])
if n == tf
else 'black'
for n in nodes],
'edgecolor':'black',
'linewidth':2,
'alpha':1},#[1 if n == tf else .1 for n in nodes]},
ckw = {})
colors,alphas,tripoints = [{} for i in range(3)]
for e in edges:
colors[e] = 'black'
alphas[e] = .2
tripoints[e] = array(pos[tf])
plot_tri = False
plot_circ = True
if plot_tri:
gd.overlay(tf_graph, pos, edges,
tripoints = tripoints,
colors = colors,
alphas = alphas)
if plot_circ:
gd.overlay(tf_graph, pos, edges,
circlepoints =circlepoints)
ax2 = fig.add_axes([.05,.05,.2,.2])
ax2.set_xticks([])
ax2.set_yticks([])
coefs = modules[tuple([tf])]['coefs']
l = len(coefs)
sx = sy = ceil(sqrt(l))
xs = mod(arange(l), sx)
ys = floor( arange(l) / sx)
cs = [colorfun([c/2]) for c in sorted(coefs)]
ss = 100
ax2.scatter(xs, ys, s = ss, color = cs)
fig.savefig(figtemplate.format('tf_{0}_net_rank{1}_{2}'.\
format(tf,max_rank,module_type)))
