def count_efms():
rates1_df, _, _ = get_concatenated_raw_data('standard')
rates2_df, _, _ = get_concatenated_raw_data('anaerobic')
rates_df = pd.concat([rates1_df, rates2_df]).drop_duplicates()
produces_biomass = (rates_df[D.R_BIOMASS] > 1e-8)
requires_oxygen = (rates_df[D.R_OXYGEN_DEPENDENT] != 0).any(axis=1)
sensitive_oxygen = (rates_df[D.R_OXYGEN_SENSITIVE].abs() > 1e-8).any(
axis=1)
print 'produces biomass = %d' % produces_biomass.sum()
print 'requires oxygen = %d' % (produces_biomass &
(requires_oxygen == True) &
(sensitive_oxygen == False)).sum()
print 'sensitive to oxygen = %d' % (produces_biomass &
(requires_oxygen == False) &
(sensitive_oxygen == True)).sum()
print 'aerobic = %d' % (produces_biomass &
(sensitive_oxygen == False)).sum()
print 'anaerobic = %d' % (produces_biomass &
(requires_oxygen == False)).sum()
print 'infeasible = %d' % (produces_biomass & (requires_oxygen == True) &
(sensitive_oxygen == True)).sum()
print 'facultative = %d' % (produces_biomass & (requires_oxygen == False) &
(sensitive_oxygen == False)).sum()
# write the data table of the main Pareto figure to a csv file:
data = pd.read_pickle(os.path.join(D.TEMP_DIR, 'standard.pkl'))
data.to_csv(os.path.join(D.TEMP_DIR, 'standard.csv'))
data = pd.read_pickle(os.path.join(D.TEMP_DIR, 'anaerobic.pkl'))
data.to_csv(os.path.join(D.TEMP_DIR, 'anaerobic.csv'))
def __init__(self, figure_data):
rates_df, _, _, _ = get_concatenated_raw_data('standard')
self.active_df = (rates_df == 0) # a boolean DataFrame of the active reactions in each EFM
self.reactions = sorted(self.active_df.columns,
key=lambda s: (int(re.findall('[rR]+(\d+)', s)[0]), s))
yield_df = figure_data['standard'][D.YIELD_L]
self.xticklabels = list(map(D.GET_REACTION_NAME, self.reactions))
self.yticklabels = list(reversed(self.xticklabels))
self.yd_double = pd.DataFrame(index=self.reactions,
columns=self.reactions, dtype=float)
self.yd_epistasis = pd.DataFrame(index=self.reactions,
columns=self.reactions, dtype=float)
max_yield = yield_df.max()
for r1 in self.reactions:
for r2 in self.reactions:
inds = self.active_df[self.active_df[r1] & self.active_df[r2]].index
self.yd_double.loc[r1, r2] = yield_df[inds].max() / max_yield
self.yd_double.fillna(0, inplace=True)
for r1 in self.reactions:
for r2 in self.reactions:
self.yd_epistasis.loc[r1, r2] = f_epi(self.yd_double, r1, r2)
self.mask1 = np.zeros_like(self.yd_double, dtype=np.bool)
self.mask1[np.triu_indices_from(self.mask1, 1)] = True
self.mask2 = np.zeros_like(self.yd_epistasis, dtype=np.bool)
self.mask2[np.triu_indices_from(self.mask2, 0)] = True
def plot_correlations(self):
# Figure that calculates the Euclidean distance between each EFM and
# the "experimental" flow, and overlays that information on the
# standard "Pareto" plot
exp_flux_df = self.fluxes_df.copy()
# remove the exchange reactions (xchg_*)
exp_flux_df = exp_flux_df.loc[
exp_flux_df.reaction_id.str.find('xchg') != 0, :]
exp_flux_df.reaction_id = exp_flux_df.reaction_id.apply(
D.FIX_REACTION_ID)
fig0, axs0 = plt.subplots(1, 2, figsize=(15, 7))
rates_df, params_df, km_df, enzyme_abundance_df = \
get_concatenated_raw_data('standard')
CORR_FLUX_L = 'correlation with exp fluxes'
LOG_LIKELIHOOD_L = 'log likelihood of flow'
figure_data = D.get_figure_data()
data = figure_data['standard']
data[CORR_FLUX_L] = rates_df.transpose().corr().loc[9999]
# calculate the likelihood of each EFM according to the measured flux
# distribution
data[LOG_LIKELIHOOD_L] = 0
joined_rates = rates_df.T
joined_rates['std'] = exp_flux_df[D.MEAS_STDEV_L]
joined_rates['std'] = joined_rates['std'].fillna(
0) + 1.0 # add a baseline stdev of 10%
for efm in data.index:
x = (joined_rates[efm] - joined_rates[9999]) / joined_rates['std']
log_likelihood = -(x**2).sum() / 2
data.loc[efm, LOG_LIKELIHOOD_L] = log_likelihood
data.loc[data[D.STRICTLY_ANAEROBIC_L],
D.GROWTH_RATE_L] = 0 # remove oxygen-sensitive EFMs
cmap = D.pareto_cmap(0.88)
D.plot_basic_pareto(data,
axs0[0],
x=D.YIELD_L,
y=D.GROWTH_RATE_L,
c=CORR_FLUX_L,
cmap=cmap,
vmin=0,
vmax=1,
linewidth=0,
s=20)
D.plot_basic_pareto(data,
axs0[1],
x=D.YIELD_L,
y=D.GROWTH_RATE_L,
c=LOG_LIKELIHOOD_L,
cmap=cmap,
linewidth=0,
s=20,
vmin=-100000,
vmax=0)
for ax in axs0:
for efm in D.efm_dict.keys():
xy = np.array(data.loc[efm,
[D.YIELD_L, D.GROWTH_RATE_L]].tolist())
xytext = xy + np.array((-1, 0.025))
ax.annotate(xy=xy,
s=D.efm_dict[efm]['label'],
xycoords='data',
xytext=xytext,
arrowprops=dict(facecolor='black',
shrink=0.05,
width=2,
headwidth=4))
ax.set_xlim(-1e-3, 1.1 * data[D.YIELD_L].max())
ax.set_ylim(-1e-3, 1.15 * data[D.GROWTH_RATE_L].max())
axs0[0].set_title('distance from measured fluxes (correlation)')
axs0[1].set_title('distance from measured fluxes (likelihood)')
fig0.tight_layout()
fig0.savefig(os.path.join(D.OUTPUT_DIR, 'Fig_flux_correlation.pdf'))
raise Exception('cannot balance reaction "%s": %s' %
(rxn, r.write_formula()))
r = Reaction(row.to_dict())
# transpose S to comply with the standard orientation for constraint-based flux models
return stoich_df.transpose()
if __name__ == '__main__':
S = read_stoichiometry()
rxn_df = S.apply(lambda r: Reaction(r.to_dict())).to_frame()
rxn_df.columns = ['Reaction']
rxn_df['dG0_prime'] = list(
rxn_df['Reaction'].apply(lambda r: r.dG0_prime()))
rates_df, _, _, _ = get_concatenated_raw_data('standard')
# keep only reactions that are in S (i.e., not external)
rates_df = rates_df[S.columns]
#%%
BOUNDS = Bounds(default_lb=1e-6, default_ub=1e-2)
mdf_data_dict = {}
for efm, row in rates_df.iterrows():
fluxes = row[row != 0]
pathway = Pathway(rxn_df.loc[fluxes.index, 'Reaction'],
fluxes.values,
rxn_df.loc[fluxes.index, 'dG0_prime'],
bounds=BOUNDS)
mdf_data_dict[efm] = pathway.calc_mdf()
print(efm, mdf_data_dict[efm].mdf)
def plot_tsne_figure(figure_data, figsize=(15, 13)):
data = figure_data['standard']
# each one of the pareto zipfiles contains the rates of all the EFMs
# so we arbitrarily chose Fig3_pareto to get them.
rates_df, _, _, _ = get_concatenated_raw_data('standard')
X = rates_df.as_matrix()
model = TSNE(n_components=2)
np.set_printoptions(suppress=True)
X_new = model.fit_transform(X)
rates_df_new = pd.DataFrame(index=rates_df.index,
columns=('t-SNE dim 1', 't-SNE dim 2'))
rates_df_new.iloc[:, 0] = X_new[:, 0]
rates_df_new.iloc[:, 1] = X_new[:, 1]
data = rates_df_new.join(data)
#%%
fig, axs = plt.subplots(3, 3, figsize=figsize, sharex=True, sharey=True)
axs = list(axs.flat)
for i, ax in enumerate(axs):
ax.annotate(chr(ord('a') + i),
xy=(0.04, 0.98),
xycoords='axes fraction',
ha='left',
va='top',
size=20)
xdata = rates_df_new.iloc[:, 0]
ydata = rates_df_new.iloc[:, 1]
axs[0].scatter(xdata, ydata, s=15, c=(0.2, 0.2, 0.7), alpha=0.3)
for efm in D.efm_dict.keys():
xy = (xdata[efm], ydata[efm])
axs[0].annotate(s=D.efm_dict[efm]['label'],
xy=xy,
xycoords='data',
xytext=(30, 5),
textcoords='offset points',
arrowprops=dict(facecolor='black',
shrink=0.05,
width=2,
headwidth=4),
ha='left',
va='bottom')
plot_parameters = [
{
'c': D.YIELD_L,
'title': 'biomass yield'
},
{
'c': D.GROWTH_RATE_L,
'title': 'growth rate'
},
{
'c': D.OXYGEN_L,
'title': 'oxygen uptake'
},
{
'c': D.ACE_L,
'title': 'acetate secretion'
},
{
'c': D.NH3_L,
'title': 'ammonia uptake'
},
{
'c': D.SUCCINATE_L,
'title': 'succinate secretion'
},
{
'c': D.ED_L,
'title': 'ED pathway'
},
{
'c': D.PPP_L,
'title': 'pentose phosphate pathway',
},
]
for i, d in enumerate(plot_parameters):
d['ax'] = axs[i + 1]
D.plot_basic_pareto(data,
x=rates_df_new.columns[0],
y=rates_df_new.columns[1],
c=d['c'],
ax=d['ax'],
cmap='copper_r',
linewidth=0.2,
s=10)
d['ax'].set_title(d['title'])
fig.tight_layout()
return fig
data = figure_data['standard']
# remove oxygen-sensitive EFMs
data.loc[data[D.STRICTLY_ANAEROBIC_L], D.GROWTH_RATE_L] = 0
D.plot_basic_pareto(data, ax2c, x=D.YIELD_L, y=D.GROWTH_RATE_L,
efm_dict=D.efm_dict,
facecolors=D.PARETO_NEUTRAL_COLOR, edgecolors='none')
ax2c.set_xlim(-1e-3, 1.1*data[D.YIELD_L].max())
ax2c.set_ylim(-1e-3, 1.15*data[D.GROWTH_RATE_L].max())
ax2c.set_title('glucose = 100 mM, O$_2$ = 3.7 mM')
fig2c.tight_layout()
fig2c.savefig(os.path.join(D.OUTPUT_DIR, 'Fig_web4.pdf'))
# %% histogram of all different EFM growth rates in a specific condition
fig5 = plt.figure(figsize=(5, 5))
ax5 = fig5.add_subplot(1, 1, 1)
efm = allocation_pie_chart(ax5, D.STD_CONC['glucoseExt'],
D.STD_CONC['oxygen'])
rates_df, full_df = get_concatenated_raw_data('sweep_glucose')
df = full_df[full_df['efm'] == efm]
v_BM = D.BIOMASS_MW * D.SECONDS_IN_HOUR * rates_df.at[efm, D.R_BIOMASS]
# make a new DataFrame where the index is the glucose concentration
# and the columns are the reactions and values are the costs.
absol = full_df[full_df['efm'] == efm].pivot(index=full_df.columns[1],
columns='reaction',
values='E_i')
fig5.savefig(os.path.join(D.OUTPUT_DIR, 'Fig_web3.pdf'))
figS6 = plot_tsne_figure(figure_data)
D.savefig(figS6, 'S6')
# %% SI Figure 7
# make bar plots for each reaction, counting how many EFMs it participates
figS7, axS7 = plt.subplots(2, 2, figsize=(15, 12))
for i, ax in enumerate(axS7.flat):
ax.annotate(chr(ord('a') + i),
xy=(0.04, 0.98),
xycoords='axes fraction',
ha='left',
va='top',
size=20)
rates1_df, _, _, _ = get_concatenated_raw_data('standard')
rates2_df, _, _, _ = get_concatenated_raw_data('anaerobic')
rates_df = pd.concat([rates1_df, rates2_df]).drop_duplicates()
reaction_counts = 100 * (rates_df.abs() > 1e-8).sum(0) / rates_df.shape[0]
plt.subplots_adjust(hspace=0.3)
reaction_counts.sort_values(inplace=True)
reaction_counts.plot(kind='bar',
ax=axS7[0, 0],
color=D.BAR_COLOR,
linewidth=0)
axS7[0, 0].set_ylim(0, 100)
axS7[0, 0].set_ylabel('\% of EFMs using this reaction')
axS7[0, 0].set_xticklabels(map(D.GET_REACTION_NAME, reaction_counts.index))