def _init_search_grid(self, surface_only=False, improvements_only=True):
"""Initialize the grid of points to be scanned within the production envelope."""
self.envelope = phenotypic_phase_plane(
self.design_space_model, self.variables, objective=self.objective, points=self.points)
intervals = self.envelope[['objective_lower_bound', 'objective_upper_bound']].copy()
intervals['objective_lower_bound'] = float_floor(intervals.objective_lower_bound, ndecimals)
intervals['objective_upper_bound'] = float_ceil(intervals.objective_upper_bound, ndecimals)
max_distance = 0.
max_interval = None
for i, (lb, ub) in intervals.iterrows():
distance = abs(ub - lb)
if distance > max_distance:
max_distance = distance
max_interval = (lb, ub)
step_size = (max_interval[1] - max_interval[0]) / (self.points - 1)
grid = list()
minimal_reference_production = self.reference_flux_ranges['lower_bound'][self.objective]
for i, row in self.envelope.iterrows():
variables = row[self.variables]
lb = row.objective_lower_bound
if improvements_only:
lb = max(lb, minimal_reference_production) + step_size
ub = row.objective_upper_bound
if not surface_only:
coordinate = lb
while coordinate < ub:
grid.append(list(variables.values) + [coordinate])
coordinate += step_size
if improvements_only and ub <= minimal_reference_production:
continue
else:
grid.append(list(variables.values) + [ub])
columns = self.variables + [self.objective]
self.grid = DataFrame(grid, columns=columns)
def _init_search_grid(self, surface_only=False, improvements_only=True):
"""Initialize the grid of points to be scanned within the production envelope."""
self.envelope = phenotypic_phase_plane(
self.design_space_model, self.variables, objective=self.objective, points=self.points)
intervals = self.envelope[['objective_lower_bound', 'objective_upper_bound']].copy()
intervals['objective_lower_bound'] = float_floor(intervals.objective_lower_bound, ndecimals)
intervals['objective_upper_bound'] = float_ceil(intervals.objective_upper_bound, ndecimals)
max_distance = 0.
max_interval = None
for i, (lb, ub) in intervals.iterrows():
distance = abs(ub - lb)
if distance > max_distance:
max_distance = distance
max_interval = (lb, ub)
step_size = (max_interval[1] - max_interval[0]) / (self.points - 1)
grid = list()
minimal_reference_production = self.reference_flux_ranges['lower_bound'][self.objective]
for i, row in self.envelope.iterrows():
variables = row[self.variables]
lb = row.objective_lower_bound
if improvements_only:
lb = max(lb, minimal_reference_production) + step_size
ub = row.objective_upper_bound
if not surface_only:
coordinate = lb
while coordinate < ub:
grid.append(list(variables.values) + [coordinate])
coordinate += step_size
if improvements_only and ub <= minimal_reference_production:
continue
else:
grid.append(list(variables.values) + [ub])
columns = self.variables + [self.objective]
self.grid = DataFrame(grid, columns=columns)
def test_float_conversions(self):
val = 1.3456
new_value = float_floor(val, 2)
assert new_value == 1.34
new_value = float_ceil(val, 2)
assert new_value == 1.35
new_value = float_floor(val, 1)
assert new_value == 1.3
new_value = float_ceil(val, 1)
assert new_value == 1.4
new_value = float_floor(val)
assert new_value == 1
new_value = float_ceil(val)
assert new_value == 2
val = 0.00000
for i in range(1, 10):
new_value = float_floor(val, i)
assert new_value == 0
new_value = float_ceil(val, i)
assert new_value == 0
def _generate_designs(cls, solutions, reference_fva, reference_fluxes):
"""
Generates strain designs for Differential FVA.
The conversion method has three scenarios:
#### 1. Knockout
Creates a ReactionKnockoutTarget.
#### 2. Flux reversal
If the new flux is negative then it should be at least the upper bound of the interval.
Otherwise it should be at least the lower bound of the interval.
#### 3. The flux increases or decreases
This table illustrates the possible combinations.
* Gap is the sign of the normalized gap between the intervals.
* Ref is the sign of the closest bound (see _closest_bound).
* Bound is the value to use
+-------------------+
| Gap | Ref | Bound |
+-----+-----+-------+
| - | - | LB |
| - | + | UB |
| + | - | UB |
| + | + | LB |
+-----+-----+-------+
Parameters
----------
solutions: pandas.Panel
The DifferentialFVA panel with all the solutions. Each DataFrame is a design.
reference_fva: pandas.DataFrame
The FVA limits for the reference strain.
reference_fluxes:
The optimal flux distribution for the reference strain.
Returns
-------
list
A list of cameo.core.strain_design.StrainDesign for each DataFrame in solutions.
"""
designs = []
for _, solution in solutions.groupby(('biomass', 'production')):
targets = []
relevant_targets = solution.loc[
(numpy.abs(solution['normalized_gaps']) > non_zero_flux_threshold) & (
numpy.logical_not(solution['excluded'])) & (
numpy.logical_not(solution['free_flux']))
]
for rid, relevant_row in relevant_targets.iterrows():
if relevant_row.KO:
targets.append(ReactionKnockoutTarget(rid))
elif relevant_row.flux_reversal:
if reference_fva['upper_bound'][rid] > 0:
targets.append(ReactionInversionTarget(rid,
value=float_ceil(relevant_row.upper_bound, ndecimals),
reference_value=reference_fluxes[rid]))
else:
targets.append(ReactionInversionTarget(rid,
value=float_floor(relevant_row.lower_bound, ndecimals),
reference_value=reference_fluxes[rid]))
else:
gap_sign = relevant_row.normalized_gaps > 0
ref_interval = reference_fva[['lower_bound', 'upper_bound']].loc[rid].values
row_interval = (relevant_row.lower_bound, relevant_row.upper_bound)
closest_bound, ref_sign = cls._closest_bound(ref_interval, row_interval)
if gap_sign ^ ref_sign:
value = float_ceil(relevant_row.upper_bound, ndecimals)
else:
value = float_floor(relevant_row.lower_bound, ndecimals)
targets.append(ReactionModulationTarget(rid,
value=value,
reference_value=reference_fva[closest_bound][rid]))
designs.append(StrainDesign(targets))
return designs
def _generate_designs(cls, solutions, reference_fva, reference_fluxes):
"""
Generates strain designs for Differential FVA.
The conversion method has three scenarios:
#### 1. Knockout
Creates a ReactionKnockoutTarget.
#### 2. Flux reversal
If the new flux is negative then it should be at least the upper bound of the interval.
Otherwise it should be at least the lower bound of the interval.
#### 3. The flux increases or decreases
This table illustrates the possible combinations.
* Gap is the sign of the normalized gap between the intervals.
* Ref is the sign of the closest bound (see _closest_bound).
* Bound is the value to use
+-------------------+
| Gap | Ref | Bound |
+-----+-----+-------+
| - | - | LB |
| - | + | UB |
| + | - | UB |
| + | + | LB |
+-----+-----+-------+
Parameters
----------
solutions: pandas.Panel
The DifferentialFVA panel with all the solutions. Each DataFrame is a design.
reference_fva: pandas.DataFrame
The FVA limits for the reference strain.
reference_fluxes:
The optimal flux distribution for the reference strain.
Returns
-------
list
A list of cameo.core.strain_design.StrainDesign for each DataFrame in solutions.
"""
designs = []
for _, solution in solutions.groupby(('biomass', 'production')):
targets = []
relevant_targets = solution.loc[
(numpy.abs(solution['normalized_gaps']) >
non_zero_flux_threshold)
& (numpy.logical_not(solution['excluded'])) &
(numpy.logical_not(solution['free_flux']))]
for rid, relevant_row in relevant_targets.iterrows():
if relevant_row.KO:
targets.append(ReactionKnockoutTarget(rid))
elif relevant_row.flux_reversal:
if reference_fva['upper_bound'][rid] > 0:
targets.append(
ReactionInversionTarget(
rid,
value=float_ceil(relevant_row.upper_bound,
ndecimals),
reference_value=reference_fluxes[rid]))
else:
targets.append(
ReactionInversionTarget(
rid,
value=float_floor(relevant_row.lower_bound,
ndecimals),
reference_value=reference_fluxes[rid]))
else:
gap_sign = relevant_row.normalized_gaps > 0
ref_interval = reference_fva[[
'lower_bound', 'upper_bound'
]].loc[rid].values
row_interval = (relevant_row.lower_bound,
relevant_row.upper_bound)
closest_bound, ref_sign = cls._closest_bound(
ref_interval, row_interval)
if gap_sign ^ ref_sign:
value = float_ceil(relevant_row.upper_bound, ndecimals)
else:
value = float_floor(relevant_row.lower_bound,
ndecimals)
targets.append(
ReactionModulationTarget(
rid,
value=value,
reference_value=reference_fva[closest_bound][rid]))
designs.append(StrainDesign(targets))
return designs
