def process_gene_knockout_solution(model, solution, simulation_method,
simulation_kwargs, biomass, target,
substrate, objective_function):
"""
Arguments
---------
model: SolverBasedModel
A constraint-based model
solution: tuple
The genes
simulation_method: function
See see cameo.flux_analysis.simulation
simulation_kwargs: dict
Keyword arguments to run the simulation method
biomass: Reaction
Cellular biomass reaction
target: Reaction
The strain design target
substrate: Reaction
The main carbon source uptake rate
objective_function: cameo.strain_design.heuristic.evolutionary.objective_functions.ObjectiveFunction
The objective function used for evaluation.
Returns
-------
list
A list with: reactions, genes, size, fva_min, fva_max, target flux, biomass flux, yield, fitness
"""
with TimeMachine() as tm:
genes = [model.genes.get_by_id(gid) for gid in solution]
reactions = find_gene_knockout_reactions(model, solution)
for reaction in reactions:
reaction.knock_out(tm)
reaction_ids = [r.id for r in reactions]
flux_dist = simulation_method(model,
reactions=objective_function.reactions,
objective=biomass,
**simulation_kwargs)
tm(do=partial(setattr, model, "objective", biomass),
undo=partial(setattr, model, "objective", model.objective))
fva = flux_variability_analysis(model,
fraction_of_optimum=0.99,
reactions=[target])
target_yield = flux_dist[target] / abs(flux_dist[substrate])
return [
tuple(reaction_ids), solution,
len(solution),
fva.lower_bound(target),
fva.upper_bound(target), flux_dist[target], flux_dist[biomass],
target_yield,
objective_function(model, flux_dist, genes)
]
def _gene_deletion(model, ids):
all_reactions = []
for g_id in ids:
all_reactions.extend(
find_gene_knockout_reactions(model,
(model.genes.get_by_id(g_id), )))
_, growth, status = _reactions_knockouts_with_restore(model, all_reactions)
return (ids, growth, status)
def visualize(self, index, map_name):
if type == REACTION_KNOCKOUT_TYPE:
knockouts = self.solutions[KNOCKOUTS][index]
else:
genes = [self.model.genes.get_by_id(g) for g in self.solutions[KNOCKOUTS][index]]
knockouts = find_gene_knockout_reactions(self.model, genes)
builder = draw_knockout_result(self.model, map_name, self.simulation_method, knockouts)
return builder.display_in_notebook()
def _gene_deletion(model, ids):
all_reactions = []
for g_id in ids:
all_reactions.extend(
find_gene_knockout_reactions(
model, (model.genes.get_by_id(g_id),)
)
)
_, growth, status = _reactions_knockouts_with_restore(model, all_reactions)
return (ids, growth, status)
def run(gene_ids):
ko_reactions = find_gene_knockout_reactions(cobra_model, gene_ids)
ko_indexes = map(cobra_model.reactions.index, ko_reactions)
# If all the reactions carry no flux, deletion will have no effect.
if no_flux_reaction_indexes.issuperset(gene_ids):
return wt_growth_rate
return moma.moma_knockout(moma_model,
moma_obj,
ko_indexes,
solver=solver,
**kwargs).f
def simulate_knockout(self, to_knockout, *args, **kwargs):
reactions = find_gene_knockout_reactions(self._model, [to_knockout])
cache = {"variables": {}, "constraints": {}, "first_run": True}
with self._model:
for reaction in reactions:
reaction.knock_out()
return self._simulation_method(
self._model, volatile=False, cache=cache, *args,
**kwargs)[cache['original_objective']]
def visualize(self, index, map_name):
if type == REACTION_KNOCKOUT_TYPE:
knockouts = self.solutions[KNOCKOUTS][index]
else:
genes = [
self.model.genes.get_by_id(g)
for g in self.solutions[KNOCKOUTS][index]
]
knockouts = find_gene_knockout_reactions(self.model, genes)
builder = draw_knockout_result(self.model, map_name,
self.simulation_method, knockouts)
return builder.display_in_notebook()
def process_gene_knockout_solution(model, solution, simulation_method, simulation_kwargs,
biomass, target, substrate, objective_function):
"""
Arguments
---------
model: cobra.Model
A constraint-based model
solution: tuple
The genes
simulation_method: function
See see cameo.flux_analysis.simulation
simulation_kwargs: dict
Keyword arguments to run the simulation method
biomass: Reaction
Cellular biomass reaction
target: Reaction
The strain design target
substrate: Reaction
The main carbon source uptake rate
objective_function: cameo.strain_design.heuristic.evolutionary.objective_functions.ObjectiveFunction
The objective function used for evaluation.
Returns
-------
list
A list with: reactions, genes, size, fva_min, fva_max, target flux, biomass flux, yield, fitness
"""
with model:
genes = [model.genes.get_by_id(gid) for gid in solution]
reactions = find_gene_knockout_reactions(model, solution)
for reaction in reactions:
reaction.knock_out()
reaction_ids = [r.id for r in reactions]
flux_dist = simulation_method(model, reactions=objective_function.reactions,
objective=biomass, **simulation_kwargs)
model.objective = biomass
fva = flux_variability_analysis(model, fraction_of_optimum=0.99, reactions=[target])
target_yield = flux_dist[target] / abs(flux_dist[substrate])
return [tuple(reaction_ids), solution, len(solution), fva.lower_bound(target), fva.upper_bound(target),
flux_dist[target], flux_dist[biomass], target_yield, objective_function(model, flux_dist, genes)]
def simulate_knockout(self, to_knockout, *args, **kwargs):
reactions = find_gene_knockout_reactions(self._model, [to_knockout])
cache = {
"variables": {},
"constraints": {},
"first_run": True
}
with TimeMachine() as tm:
for reaction in reactions:
reaction.knock_out(tm)
return self._simulation_method(self._model,
volatile=False,
cache=cache,
*args,
**kwargs)[cache['original_objective']]
def __call__(self, individual):
"""
Parameters
----------
individual : list
a list of integers
Returns
-------
list
[knockouts, decoded representation]
"""
genes = [self.model.genes.get_by_id(self.representation[index]) for index in individual]
reactions = find_gene_knockout_reactions(self.model, genes)
return [reactions, genes]
def gene_knockout_growth(
gene_id,
model,
threshold=10 ** -6,
simulation_method=fba,
normalize=True,
biomass=None,
biomass_flux=None,
*args,
**kwargs
):
if biomass_flux is None:
s = model.solve()
biomass_flux = s.f
if "reference" not in kwargs:
kwargs["reference"] = s.x_dict
gene = model.genes.get_by_id(gene_id)
knockouts = find_gene_knockout_reactions(model, [gene])
tm = TimeMachine()
for reaction in knockouts:
tm(
do=partial(setattr, reaction, "lower_bound", 0),
undo=partial(setattr, reaction, "lower_bound", reaction.lower_bound),
)
tm(
do=partial(setattr, reaction, "upper_bound", 0),
undo=partial(setattr, reaction, "upper_bound", reaction.upper_bound),
)
try:
s = simulation_method(model, *args, **kwargs)
f = s.get_primal_by_id(biomass)
if f >= threshold:
if normalize:
f = f / biomass_flux
else:
f = 0
except SolveError:
f = float("nan")
finally:
tm.reset()
return f
def __call__(self, individual):
"""
Parameters
----------
individual : list
a list of integers
Returns
-------
list
[knockouts, decoded representation]
"""
genes = [
self.model.genes.get_by_id(self.representation[index])
for index in individual
]
reactions = find_gene_knockout_reactions(self.model, genes)
return [reactions, genes]
def gene_knockout_growth(gene_id,
model,
threshold=10**-6,
simulation_method=fba,
normalize=True,
biomass=None,
biomass_flux=None,
*args,
**kwargs):
if biomass_flux is None:
s = model.solve()
biomass_flux = s.f
if 'reference' not in kwargs:
kwargs['reference'] = s.x_dict
gene = model.genes.get_by_id(gene_id)
knockouts = find_gene_knockout_reactions(model, [gene])
tm = TimeMachine()
for reaction in knockouts:
tm(do=partial(setattr, reaction, 'lower_bound', 0),
undo=partial(setattr, reaction, 'lower_bound',
reaction.lower_bound))
tm(do=partial(setattr, reaction, 'upper_bound', 0),
undo=partial(setattr, reaction, 'upper_bound',
reaction.upper_bound))
try:
s = simulation_method(model, *args, **kwargs)
f = s.get_primal_by_id(biomass)
if f >= threshold:
if normalize:
f = f / biomass_flux
else:
f = 0
except SolveError:
f = float('nan')
finally:
tm.reset()
return f
def __call__(self, individual, flat=False):
"""
Parameters
----------
individual: list
a list of integers
flat: bool
if True, returns strings. Otherwise returns Gene
Returns
-------
list
[knockouts, decoded representation]
"""
genes = [self.model.genes.get_by_id(self.representation[index]) for index in individual]
reactions = find_gene_knockout_reactions(self.model, genes)
if flat:
return [tuple(r.id for r in reactions), tuple(g.id for g in genes)]
return [tuple(reactions), tuple(genes)]
def find_essential_genes_reactions(self):
""" If the model has genes finds the reactions associated with the essential genes.
Searches essential genes if hasn't done before.
"""
errors = []
try:
if self.__essential_genes is not None:
# Dict with reaction as key and list of genes as value
reactions = {}
for gene in self.__essential_genes:
reactions_knock = find_gene_knockout_reactions(
self.__cobra_model, [gene])
for reaction in reactions_knock:
if reaction in reactions:
reactions[reaction].append(gene)
else:
reactions[reaction] = [gene]
self.__essential_genes_reactions = reactions
except Exception as error:
self.__essential_genes_reactions = {}
errors.append(str(error))
return errors
def knock_out(self, time_machine=None):
"""Knockout gene by setting all its affected reactions' bounds to zero.
Parameters
----------
time_machine = TimeMachine
A time TimeMachine instance can be provided to undo the knockout eventually.
Returns
-------
None
"""
from cobra.manipulation.delete import find_gene_knockout_reactions
for reaction in find_gene_knockout_reactions(self.model, [self]):
def _(reaction, lb, ub):
reaction.upper_bound = ub
reaction.lower_bound = lb
if time_machine is not None:
time_machine(do=reaction.knock_out, undo=partial(_, reaction, reaction.lower_bound, reaction.upper_bound))
else:
reaction.knock_out()
def _double_gene_deletion_fba(cobra_model,
gene_ids1,
gene_ids2,
gene_id_to_result,
solver,
number_of_processes=None,
zero_cutoff=1e-12,
wt_growth_rate=None,
no_flux_reaction_indexes=set(),
**kwargs):
"""compute double gene deletions using fba
cobra_model: model
gene_ids1, gene_ids2: lists of id's to be knocked out
gene_id_to_result: maps each gene identifier to the entry in
the result matrix
no_flux_reaction_indexes: set of indexes for reactions in the model
which carry no flux in an optimal solution. For deletions only in
this set, the result will beset to wt_growth_rate.
returns an upper triangular square matrix
"""
# Because each gene reaction rule will be evaluated multiple times
# the reaction has multiple associated genes being deleted, compiling
# the gene reaction rules ahead of time increases efficiency greatly.
compiled_rules = get_compiled_gene_reaction_rules(cobra_model)
n_results = len(gene_id_to_result)
results = numpy.empty((n_results, n_results))
results.fill(numpy.nan)
if number_of_processes == 1: # explicitly disable multiprocessing
PoolClass = CobraDeletionMockPool
else:
PoolClass = CobraDeletionPool
with PoolClass(cobra_model,
n_processes=number_of_processes,
solver=solver,
**kwargs) as pool:
# precompute all single deletions in the pool and store them along
# the diagonal
for gene_id, gene_result_index in iteritems(gene_id_to_result):
ko_reactions = find_gene_knockout_reactions(
cobra_model, (cobra_model.genes.get_by_id(gene_id), ))
ko_indexes = [cobra_model.reactions.index(i) for i in ko_reactions]
pool.submit(ko_indexes, label=gene_result_index)
for result_index, value in pool.receive_all():
# if singly lethal, set everything in row and column to 0
value = value if abs(value) > zero_cutoff else 0.
if value == 0.:
results[result_index, :] = 0.
results[:, result_index] = 0.
else: # only the diagonal needs to be set
results[result_index, result_index] = value
# Run double knockouts in the upper triangle
index_selector = yield_upper_tria_indexes(gene_ids1, gene_ids2,
gene_id_to_result)
for result_index, (gene1, gene2) in index_selector:
# if singly lethal the results have already been set
if results[result_index] == 0:
continue
ko_reactions = find_gene_knockout_reactions(
cobra_model, (gene1, gene2), compiled_rules)
ko_indexes = [cobra_model.reactions.index(i) for i in ko_reactions]
# if all removed gene indexes carry no flux
if len(set(ko_indexes) - no_flux_reaction_indexes) == 0:
results[result_index] = wt_growth_rate
continue
pool.submit(ko_indexes, label=result_index)
for result in pool.receive_all():
value = result[1]
if value < zero_cutoff:
value = 0
results[result[0]] = value
return results
def single_gene_deletion_fba(cobra_model, gene_list, solver=None,
**solver_args):
"""Sequentially knocks out each gene in a model using FBA.
Not supposed to be called directly use
`single_reactions_deletion(..., method="fba")` instead.
Parameters
----------
gene_list : iterable
List of gene IDs or cobra.Reaction.
solver: str, optional
The name of the solver to be used.
Returns
-------
tuple of dicts
A tuple ({reaction_id: growth_rate}, {reaction_id: status})
"""
legacy = False
if solver is None:
solver = cobra_model.solver
elif "optlang-" in solver:
solver = solvers.interface_to_str(solver)
solver = solvers.solvers[solver]
else:
legacy = True
solver = legacy_solvers.solver_dict[solver]
lp = solver.create_problem(cobra_model)
growth_rate_dict = {}
status_dict = {}
if not legacy:
with cobra_model as m:
m.solver = solver
for gene in gene_list:
ko = find_gene_knockout_reactions(cobra_model, [gene])
with m:
for reaction in ko:
reaction.bounds = (0.0, 0.0)
m.solver.optimize()
status = m.solver.status
status_dict[gene.id] = status
growth_rate_dict[gene.id] = m.solver.objective.value if \
status == OPTIMAL else 0.
else:
for gene in gene_list:
old_bounds = {}
for reaction in find_gene_knockout_reactions(cobra_model, [gene]):
index = cobra_model.reactions.index(reaction)
old_bounds[index] = reaction.bounds
solver.change_variable_bounds(lp, index, 0., 0.)
solver.solve_problem(lp, **solver_args)
# get the status and growth rate
status = solver.get_status(lp)
status_dict[gene.id] = status
growth_rate = solver.get_objective_value(lp) \
if status == "optimal" else 0.
growth_rate_dict[gene.id] = growth_rate
# reset the problem
for index, bounds in iteritems(old_bounds):
solver.change_variable_bounds(lp, index, bounds[0], bounds[1])
return growth_rate_dict, status_dict
def single_gene_deletion_moma(cobra_model, gene_list, solver=None,
**solver_args):
"""Sequentially knocks out each gene in a model using MOMA.
Not supposed to be called directly use
`single_reactions_deletion(..., method="moma")` instead.
Parameters
----------
gene_list : iterable
List of gene IDs or cobra.Reaction.
solver: str, optional
The name of the solver to be used.
Returns
-------
tuple of dicts
A tuple ({reaction_id: growth_rate}, {reaction_id: status})
"""
if moma is None:
raise RuntimeError("scipy required for moma")
legacy = False
if solver is None:
solver = cobra_model.solver
elif "optlang-" in solver:
solver = solvers.interface_to_str(solver)
solver = solvers.solvers[solver]
else:
legacy = True
solver = legacy_solvers.solver_dict[solver]
moma_model, moma_objective = moma.create_euclidian_moma_model(
cobra_model)
growth_rate_dict = {}
status_dict = {}
if not legacy:
with cobra_model as m:
m.solver = solver
moma.add_moma(m)
for gene in gene_list:
ko = find_gene_knockout_reactions(cobra_model, [gene])
with m:
for reaction in ko:
reaction.bounds = (0.0, 0.0)
m.solver.optimize()
status = m.solver.status
status_dict[gene.id] = status
if status == "optimal":
growth = m.variables.moma_old_objective.primal
else:
growth = 0.0
growth_rate_dict[gene.id] = growth
else:
for gene in gene_list:
delete_model_genes(moma_model, [gene.id])
solution = moma.solve_moma_model(moma_model, moma_objective,
solver=solver, **solver_args)
status_dict[gene.id] = solution.status
growth_rate_dict[gene.id] = solution.f
undelete_model_genes(moma_model)
return growth_rate_dict, status_dict
def test_gene_knockout_decoder(self):
decoder = GeneKnockoutDecoder([g.id for g in self.model.genes], self.model)
reactions1, genes = decoder([1, 2, 3, 4])
reactions2 = find_gene_knockout_reactions(self.model, genes)
self.assertTrue(sorted(reactions1, key=lambda x: x.id) == sorted(reactions2, key=lambda x: x.id))
def test_gene_knockout_decoder(self):
decoder = GeneKnockoutDecoder([g.id for g in self.model.genes], self.model)
reactions1, genes = decoder([1, 2, 3, 4])
reactions2 = find_gene_knockout_reactions(self.model, genes)
self.assertTrue(sorted(reactions1, key=lambda x: x.id) == sorted(reactions2, key=lambda x: x.id))