def plot_gene_prediction(gene,
match,
data,
sim_der,
col_names,
ensembl_to_hgnc,
ax=None,
**kwargs):
dep = DEPlot()
matching_sim = match.loc[match.train_gene == gene,
'net'].astype(str).values
pred_lfc = sim_der.coefficients.loc[matching_sim]
baseline = data.loc[gene, 'wt'].groupby('time').mean().values
random = sim_der.coefficients + baseline
random.columns = col_names
random = random.unstack()
random.index = random.index.set_names(['time', 'replicate'])
random.name = 'random'
random = pd.concat([random], keys=['random'], names=['condition'])
pred = pred_lfc + baseline
pred.columns = col_names
pred = pred.unstack()
pred.index = pred.index.set_names(['time', 'replicate'])
pred.name = 'predicted'
pred = pd.concat([pred], keys=['predicted'], names=['condition'])
true = data.loc[gene, ['ki']]
pred = pred.reorder_levels(true.index.names)
random = random.reorder_levels(true.index.names)
ts_data = pd.concat([true, pred, random])
ts_data.name = gene
ax = dep.tsplot(ts_data, scatter=False, ax=ax, legend=False, **kwargs)
ax.set_title(ensembl_to_hgnc.loc[gene, 'hgnc_symbol'])
ax.set_ylabel('')
def display_sim(network,
stim,
perturbation,
times,
directory,
exp_condition='ko',
ctrl_condition='wt',
node=None):
data_dir = '{}/{}/{}/'.format(directory, network, stim)
network_structure = "{}/{}/{}_goldstandard_signed.tsv".format(
directory, network, network)
ctrl_gsr = GnwSimResults(
data_dir,
network,
ctrl_condition,
sim_suffix='dream4_timeseries.tsv',
perturb_suffix="dream4_timeseries_perturbations.tsv")
exp_gsr = GnwSimResults(
data_dir,
network,
exp_condition,
sim_suffix='dream4_timeseries.tsv',
perturb_suffix="dream4_timeseries_perturbations.tsv")
data = pd.concat([ctrl_gsr.data,
exp_gsr.data]).T.sort_index(axis=0).sort_index(axis=1)
if node:
dep = DEPlot()
idx = pd.IndexSlice
dep.tsplot(data.loc[node, idx[:, :, perturbation, times]],
subgroup='Time',
no_fill_legend=True)
plt.tight_layout()
return
dg = get_graph(network_structure)
titles = ["x", "y", "PI3K"]
mapping = {'G': "PI3k"}
dg = nx.relabel_nodes(dg, mapping)
draw_results(np.log2(data + 1), perturbation, titles, times=times, g=dg)
plt.tight_layout()
def plot_genes(sub_spec, leg_spec, net_spec, matches, dde, sim_dea, tx_to_gene,
net_data, ts):
# genes = [top.index[29], ts.sort_values('percent', ascending=False).index[0],
# ts.sort_values('percent', ascending=False).index[30]]
# genes = ts.index[np.random.randint(0, len(top), size=3)]
# genes = top.sort_values('grouped_e').index[:3]
genes = ['ENSG00000117289', 'ENSG00000213626', 'ENSG00000170044']
w_ratios = [1] * len(genes) + [0.1, 0.1]
cols = len(genes) + 2
gs_top = gridspec.GridSpecFromSubplotSpec(2,
cols,
subplot_spec=sub_spec,
width_ratios=w_ratios,
wspace=0.5)
gs_bottom = gridspec.GridSpecFromSubplotSpec(1,
4,
subplot_spec=leg_spec,
width_ratios=[1, 1, 0.1, 0.1],
wspace=0.5)
gs_mid = gridspec.GridSpecFromSubplotSpec(1,
cols,
subplot_spec=net_spec,
width_ratios=w_ratios,
wspace=0.5)
train_conditions = list(dde.training.values())
dep = DEPlot()
greys = ['0.1', '0.7', '0.4', '0.1', '0.7']
conditions = ['wt', 'ko', 'ki', 'predicted', 'random']
colors = {c: idx for c, idx in zip(conditions, greys)}
# colors['random'] = '0.7'
three_col = ['#AA0000', '#BBBBBB', '#0000AA']
new_map = LinearSegmentedColormap.from_list('custom', three_col)
edge_cmap = new_map
pie_colors = ['#0F8554', '#E17C05', '0.2', '0.7']
train_markers = ['o', '^']
test_markers = ['d', 'X', 's']
plot_times = dde.times.copy()
plot_times.remove(15)
for idx, gene in enumerate(genes):
with sns.axes_style("whitegrid"):
train_ax = plt.subplot(gs_top[0, idx])
pred_ax = plt.subplot(gs_top[1, idx])
dep.tsplot(dde.dea.data.loc[gene, train_conditions],
ax=train_ax,
legend=False,
no_fill_legend=True,
color_dict=colors,
scatter=False,
markers=train_markers)
train_ax.set_title(tx_to_gene.loc[gene, 'hgnc_symbol'])
train_ax.set_ylabel('log2(counts)')
train_ax.yaxis.set_major_formatter(FormatStrFormatter('%.0f'))
train_ax.set_xlabel('')
train_ax.set_xlim(min(plot_times), max(plot_times))
train_ax.set_xticks(plot_times)
train_ax.set_xticklabels([])
plot_gene_prediction(gene,
matches,
dde.dea.data,
sim_dea.results['ki-wt'],
sim_dea.times,
tx_to_gene,
ax=pred_ax,
no_fill_legend=True,
color_dict=colors,
markers=test_markers)
pred_ax.set_xlim(min(plot_times), max(plot_times))
pred_ax.set_xticks(plot_times)
pred_ax.set_xticklabels(plot_times, rotation=90)
pred_ax.set_ylabel('log2(counts)')
pred_ax.yaxis.set_major_formatter(FormatStrFormatter('%.0f'))
pred_ax.set_title('')
if idx != 0:
train_ax.set_ylabel('')
pred_ax.set_ylabel('')
net_ax = plt.subplot(gs_mid[0, idx])
labels = ['G', 'x', 'y']
logics = ['_multiplicative', '_linear', '']
node_info = {}
models = net_data.loc[matches[matches.train_gene == gene]
['net'].values]
for node in labels:
cur_dict = {}
counts = Counter(models['{}_logic'.format(node)])
cur_dict['fracs'] = [
counts['{}{}'.format(node, log)] for log in logics
]
no_in = sum(models['{}_in'.format(node)] == 0)
cur_dict['fracs'][-1] -= no_in
cur_dict['fracs'].append(no_in)
node_info[node] = cur_dict
plot_net(net_ax, node_info, models, labels, edge_cmap, pie_colors)
# Add the net legend
if idx == len(genes) - 1:
leg = net_ax.legend(['AND', 'OR', 'Single input', 'No input'],
loc='center left',
bbox_to_anchor=(1, 0.5),
handletextpad=0.5,
frameon=False,
handlelength=1)
leg.set_title("Node regulation", prop={'size': 24})
# Add the legends
labels = {'wt': 'Wildtype', 'ko': 'PI3K KO', 'ki': 'PI3K KI'}
for line in train_ax.get_lines() + pred_ax.get_lines():
label = line.get_label()
try:
new = labels[label]
except KeyError:
new = label.capitalize()
line.set_label(new)
train_leg = train_ax.legend(loc='center left',
bbox_to_anchor=(1, 0.5),
frameon=False,
handlelength=1)
pred_leg = pred_ax.legend(loc='center left',
bbox_to_anchor=(1, 0.5),
frameon=False,
handlelength=1)
train_leg.set_title('Training', prop={'size': 28, 'weight': 'bold'})
pred_leg.set_title('Testing', prop={'size': 28, 'weight': 'bold'})
# Add arrow
sns.set_style(None)
arrow_ax = plt.subplot(gs_bottom[0])
despine(arrow_ax)
detick(arrow_ax)
# Scale arrow width
fig = arrow_ax.get_figure()
bbox = arrow_ax.get_window_extent().transformed(
fig.dpi_scale_trans.inverted())
width, height = bbox.width, bbox.height
height *= fig.dpi / 2
astyle = ArrowStyle('wedge', tail_width=height, shrink_factor=0.5)
fa = FancyArrowPatch(posA=[0, 0.5],
posB=[1, 0.5],
arrowstyle=astyle,
lw=0,
color='k')
arrow_ax.add_artist(fa)
arrow_ax.set_title('fraction of models \n edge exists')
arrow_ax.set_xticks([0, 1])
arrow_ax.set_xticklabels(['100%', '0%'])
# Add colorbar
norm = mpl.colors.Normalize(vmin=-1, vmax=1)
cbar_ax = plt.subplot(gs_bottom[1])
cb1 = mpl.colorbar.ColorbarBase(cbar_ax,
cmap=edge_cmap,
norm=norm,
orientation='horizontal')
cb1.set_ticks([-1, 0, 1])
cbar_ax.set_title('Average edge sign')
# a = 0
# unique = 0
# for gene, data in grouped:
# unique += 1
# if gene in true_dde_genes.index:
# print(data)
# a += 1
# print(a)
# print(true_dde_genes.shape)
# print(unique)
# sys.exit()
# print(all)
dep = DEPlot()
dep.tsplot(dea.voom_data.loc['ENSG00000004799',
contrast.replace('ki', 'ko').split('-')],
legend=False)
plt.tight_layout()
dep.tsplot(true_dea.voom_data.loc['ENSG00000004799',
contrast.split('-')],
legend=False)
plt.tight_layout()
# Display results
display_sim(1995, -1, times, "../data/motif_library/gnw_networks/")
display_sim(1995,
-1,
times,
"../data/motif_library/gnw_networks/",
exp_condition='ki')
plt.savefig(
'/Users/jfinkle/Dropbox/Justin Finkle, shared/Media/Sprouty/images/row_max_normalized_heatmap.png',
fmt='png',
dpi=600)
#
#
sys.exit()
# Volcano Plot
x = dea.results['KO-WT'].top_table(coef=1, use_fstat=False)
# dep.volcano_plot(x, top_n=5, show_labels=True)
# Time Series Plot
x = dea.data.loc['IVL']
dep.tsplot(x)
plt.show()
sys.exit()
x = dea.results['KO_ts-WT_ts']
y = dea.results['KO-WT']
gene = 'CXCL1'
# print(x.discrete_clusters[(x.discrete_clusters['Cluster'] != '(0, 0, 0, 0)') & (y.discrete['KO_0-WT_0'] == 0)])
print(x.continuous.head())
print(dea.results['(KO-WT)_ts'].continuous.head())
p_data = dea.data.loc[gene]
dep.tsplot(p_data)
plt.show()
sep=' ')
gene_names = gene_names[['id',
'hgnc_symbol']].set_index('id').drop_duplicates()
p_thresh = 0.001
scores = p_results.loc[(der.top_table()['adj_pval'] < p_thresh)
& (p_results['p_value'] < p_thresh)]
# Remove clusters that have no dynamic DE (i.e. all 1, -1, 0)
interesting = scores.loc[
scores.Cluster.apply(ast.literal_eval).apply(set).apply(len) > 1]
# print(interesting.sort_values(['score'], ascending=False).head(100))
c = (interesting[interesting.Cluster == '(0, 0, 0, 0, -1, -1)'].sort_values(
'score', ascending=False))
print(c)
dep = DEPlot()
dep.tsplot(dea.data.loc["ENSG00000186187", ['ko', 'wt']], legend=False)
plt.title('ZNRF1')
plt.tight_layout()
plt.show()
sys.exit()
for gene in c.index:
dep.tsplot(dea.data.loc[gene, ['pten', 'wt']], legend=False)
plt.tight_layout()
plt.show()
sys.exit()
# Heatmap of expression
de_data = (der.top_table().iloc[:, :6]) #.multiply(der.p_value < p_thresh)
sort_idx = interesting.sort_values(['Cluster', 'score'],
ascending=False).index.values
hm_data = de_data.loc[sort_idx]