'''Define the InstFitter class.'''
from functools import partial
from collections.abc import Iterable
from copy import deepcopy
from math import ceil
import numpy as np
from multiprocessing import Pool
from ..io.inputs import read_measurements_from_file, read_initial_values
from ..io.results import InstFitResults, InstFitMCResults
from .inst_simulate import InstSimulator
from .fit import Fitter
from ..solver.nlpsolver import InstMFAModel
from ..utils.progress import Progress
[docs]
class InstFitter(Fitter, InstSimulator):
'''
Estimated fluxes are in the unit of umol/gCDW/s if concentrations in the unit of
umol/gCDW and timepoints in the unit of s.
'''
[docs]
def set_measured_MDVs(self, fragmentid, timepoints, means, sds):
'''
Set measured MDVs at various timepoints.
Parameters
----------
fragmentid: str
Metabolite ID + '_' + atom NOs, e.g., 'Glu_12345'.
timepoints: float or list of float
Timepoint(s).
means: array or list of array
Mean of measured MDV(s). len(means) should be equal to len(timepoints).
sds: array or list of array
Standard deviation of measured MDV(s). len(sds) should be equal to len(timepoints).
'''
if not isinstance(timepoints, Iterable):
timepoints = [timepoints]
means = [means]
sds = [sds]
for timepoint, mean, sd in zip(timepoints, means, sds):
self.model.measured_inst_MDVs.setdefault(fragmentid, {})[timepoint] = [
np.array(mean),
np.array(sd)
]
if self.contexts:
context = self.contexts[-1]
context.add_undo(partial(self._unset_measured_MDVs, {fragmentid: timepoints}))
[docs]
def set_measured_MDVs_from_file(self, file):
'''
Read measured MDVs at various timepoints from file.
Parameters
----------
file: file path
Path of tsv or excel file with fields "fragment_ID", "time", "mean" and "sd".
"fragment_ID" is metabolite ID + '_' + atom NOs, e.g., 'Glu_12345';
"time" is timepoint when MDVs are measured (while some timepoints could be missing);
"mean" and "sd" are the mean and standard deviation of MDV with element seperated by ','.
Header line starts with "#", and will be skiped.
'''
measMDVs = read_measurements_from_file(file, inst_data = True)
fragmentid_tpoints = {}
for [emuid, timepoint], [mean, sd] in measMDVs.iterrows():
timepoint = float(timepoint)
self.model.measured_inst_MDVs.setdefault(emuid, {})[timepoint] = [
np.array(list(map(float, mean.split(',')))),
np.array(list(map(float, sd.split(','))))
]
fragmentid_tpoints.setdefault(emuid, []).append(timepoint)
if self.contexts:
context = self.contexts[-1]
context.add_undo(partial(self._unset_measured_MDVs, fragmentid_tpoints))
[docs]
def _unset_measured_MDVs(self, fragmentid_tpoints):
'''
Parameters
----------
fragmentid_tpoints: dict
measured MDV ID => list of timepoints.
'''
for fragmentid, timepoints in fragmentid_tpoints.items():
if fragmentid in self.model.measured_inst_MDVs:
for timepoint in timepoints:
if timepoint in self.model.measured_inst_MDVs[fragmentid]:
self.model.measured_inst_MDVs[fragmentid].pop(timepoint)
if not self.model.measured_inst_MDVs[fragmentid]:
self.model.measured_inst_MDVs.pop(fragmentid)
[docs]
def set_concentration_bounds(self, metabid, bounds):
'''
Set lower and upper bounds for concentration in unit of umol/gCDW.
Parameters
----------
metabid: str or 'all'
Metabolite ID. If 'all', all concentrations will be set to the range.
bounds: 2-list
[lower bound, upper bound]. Lower bound is not allow to equal upper bound.
'''
bounds = list(map(float, bounds))
metabids = []
if bounds[0] < bounds[1]:
if metabid == 'all':
for metabid in self.model.metabolites:
self.model.concentrations_bounds[metabid] = [max(0.0, bounds[0]), bounds[1]]
metabids.append(metabid)
elif metabid in self.model.metabolites:
self.model.concentrations_bounds[metabid] = [max(0.0, bounds[0]), bounds[1]]
metabids = [metabid]
else:
raise ValueError(f'concentration range set to nonexistent metabolite {metabid}')
else:
raise ValueError('concentration lower bound should be less than upper bound')
if self.contexts:
context = self.contexts[-1]
context.add_undo(partial(self._unset_concentration_bounds, metabids))
[docs]
def _set_default_concentration_bounds(self):
'''
This method assign bounds of [0.01, 100] (umol/gCDW) for concentrations
not set by set_concentration_bounds
'''
defBnds = [0.01, 100]
metabids = []
for metabid in self.model.metabolites:
if metabid not in self.model.concentrations_bounds:
self.model.concentrations_bounds[metabid] = defBnds
metabids.append(metabid)
if self.contexts:
context = self.contexts[-1]
context.add_undo(partial(self._unset_concentration_bounds, metabids))
[docs]
def _unset_concentration_bounds(self, metabids):
'''
Parameters
----------
metabids: str or list of str
Metabolite ID(s).
'''
if not isinstance(metabids, Iterable):
metabids = [metabids]
for metabid in metabids:
if metabid in self.model.concentrations_bounds:
self.model.concentrations_bounds.pop(metabid)
[docs]
def _decompose_network(self, n_jobs):
'''
Parameters
----------
n_jobs: int
# of jobs to run in parallel.
'''
if not self.model.measured_inst_MDVs:
raise ValueError('call set_measured_MDV or set_measured_MDVs_from_file first')
if not self.model.EAMs:
if n_jobs <= 0:
raise ValueError('n_jobs should be a positive value')
else:
self.model.target_EMUs = list(self.model.measured_inst_MDVs.keys())
metabids = []
atom_nos = []
for emuid in self.model.target_EMUs:
metabid, atomNOs = emuid.split('_')
metabids.append(metabid)
atom_nos.append(atomNOs)
EAMs = self.model._decompose_network(
metabids, atom_nos,
lump = False,
n_jobs = n_jobs
)
for size, EAM in EAMs.items():
self.model.EAMs[size] = EAM
if self.contexts:
context = self.contexts[-1]
context.add_undo(self._unset_decomposition)
[docs]
def _set_timepoints(self):
if not self.model.timepoints:
self.calculator._set_timepoints()
if self.contexts:
context = self.contexts[-1]
context.add_undo(self._unset_timepoints)
[docs]
def _calculate_matrix_Ms_derivatives_p(self):
if not self.model.matrix_Ms_der_p:
self.calculator._calculate_matrix_Ms_derivatives_p()
if self.contexts:
context = self.contexts[-1]
context.add_undo(self._unset_matrix_Ms_derivatives_p)
[docs]
def _unset_matrix_Ms_derivatives_p(self):
self.model.matrix_Ms_der_p.clear()
[docs]
def _calculate_measured_inst_MDVs_inversed_covariance_matrix(self):
if not self.model.measured_inst_MDVs_inv_cov:
self.calculator._calculate_measured_inst_MDVs_inversed_covariance_matrix()
if self.contexts:
context = self.contexts[-1]
context.add_undo(self._unset_measured_inst_MDVs_inversed_covariance_matrix)
[docs]
def _unset_measured_inst_MDVs_inversed_covariance_matrix(self):
self.model.measured_inst_MDVs_inv_cov = None
[docs]
def _calculate_initial_matrix_Xs_derivatives_p(self):
if not self.model.initial_matrix_Xs_der_p:
self.calculator._calculate_initial_matrix_Xs_derivatives_p()
if self.contexts:
context = self.contexts[-1]
context.add_undo(self._unset_initial_matrix_Xs_derivatives_p)
[docs]
def _unset_initial_matrix_Xs_derivatives_p(self):
self.model.initial_matrix_Xs_der_p.clear()
[docs]
def _calculate_initial_matrix_Ys_derivatives_p(self):
if not self.model.initial_matrix_Ys_der_p:
self.calculator._calculate_initial_matrix_Ys_derivatives_p()
if self.contexts:
context = self.contexts[-1]
context.add_undo(self._unset_initial_matrix_Ys_derivatives_p)
[docs]
def _unset_initial_matrix_Ys_derivatives_p(self):
self.model.initial_matrix_Ys_der_p.clear()
[docs]
def _estimate_concentrations_range(self):
if not self.model.concentrations_range:
for metabid in self.model.concids:
self.model.concentrations_range[metabid] = self.model.concentrations_bounds[metabid]
if self.contexts:
context = self.contexts[-1]
context.add_undo(partial(self._unset_concentrations_range, self.model.concids))
[docs]
def _unset_concentrations_range(self, metabids):
'''
Parameters
----------
metabids: str or list of str
Metabolite ID(s).
'''
if not isinstance(metabids, Iterable):
metabids = [metabids]
for metabid in metabids:
if metabid in self.model.concentrations_range:
self.model.concentrations_range.pop(metabid)
[docs]
def prepare(self, dilution_from = None, n_jobs = 1):
'''
Parameters
----------
dilution_from: str or list of str
ID(s) of unlabeled (inactive) metabolite leading to dilution effect.
These metabolites have zero stoichiometric coefficients in reaction network.
n_jobs: int
If n_jobs > 1, decomposition job will run in parallel.
'''
self._decompose_network(n_jobs)
self._set_timepoints()
self._calculate_null_space()
self._calculate_transform_matrix()
self._lambdify_matrix_As_and_Bs()
self._calculate_matrix_As_and_Bs_derivatives_p('inst', n_jobs)
self._lambdify_matrix_Ms()
self._calculate_matrix_Ms_derivatives_p()
self._calculate_substrate_MDVs(dilution_from)
self._calculate_substrate_MDV_derivatives_p('inst', dilution_from)
self._calculate_measured_inst_MDVs_inversed_covariance_matrix()
self._calculate_measured_fluxes_inversed_covariance_matrix()
self._calculate_measured_fluxes_derivative_p('inst')
self._calculate_initial_matrix_Xs()
self._calculate_initial_matrix_Ys()
self._calculate_initial_matrix_Xs_derivatives_p()
self._calculate_initial_matrix_Ys_derivatives_p()
self._set_default_flux_bounds()
self._estimate_fluxes_range(self.model.unbalanced_metabolites)
self._set_default_concentration_bounds()
self._estimate_concentrations_range()
[docs]
def _check_dependencies(self, fit_measured_fluxes):
'''
Parameters
----------
fit_measured_fluxes: bool
Whether to fit measured fluxes.
'''
if not self.model.net_fluxes_bounds:
raise ValueError('call set_flux_bounds first')
if not self.model.concentrations_bounds:
raise ValueError('call set_concentration_bounds first')
if not self.model.measured_inst_MDVs:
raise ValueError('call call set_measured_MDV or set_measured_MDVs_from_file first')
if not self.model.measured_fluxes:
raise ValueError('call set_measured_flux or set_measured_fluxes_from_file first')
if not self.model.labeling_strategy:
raise ValueError('call labeling_strategy first')
checklist = [
not self.model.target_EMUs,
self.model.transform_matrix is None,
self.model.null_space is None,
self.model.measured_inst_MDVs_inv_cov is None,
not self.model.matrix_As,
not self.model.matrix_Bs,
not self.model.matrix_Ms,
not self.model.matrix_As_der_p,
not self.model.matrix_Bs_der_p,
not self.model.matrix_Ms_der_p,
not self.model.substrate_MDVs,
not self.model.substrate_MDVs_der_p,
self.model.measured_fluxes_der_p is None,
not self.model.initial_matrix_Xs,
not self.model.initial_matrix_Ys,
not self.model.initial_matrix_Xs_der_p,
not self.model.initial_matrix_Ys_der_p,
not self.model.timepoints
]
if fit_measured_fluxes:
checklist.append(self.model.measured_fluxes_inv_cov is None)
if any(checklist):
raise ValueError('call prepare first')
[docs]
def solve(
self,
fit_measured_fluxes = True,
ini_fluxes = None,
ini_concs = None,
solver = 'slsqp',
tol = 1e-6,
max_iters = 400,
show_progress = True
):
'''
Parameters
----------
fit_measured_fluxes: bool
Whether to fit measured fluxes.
ini_fluxes: ser or file in .tsv or .xlsx
Initial values of net fluxes.
ini_concs: ser or file in .tsv or .xlsx
Initial values of concentrations.
solvor: {"slsqp", "ralg"}
* If "slsqp", scipy.optimize.minimze will be used.
* If "ralg", openopt NLP solver will be used.
tol: float
Tolerance for termination.
max_iters: int
Maximum # of iterations.
show_progress: bool
Whether to show the progress bar.
'''
self._check_dependencies(fit_measured_fluxes)
if ini_fluxes is not None:
iniFluxes = read_initial_values(ini_fluxes, self.model.netfluxids)
else:
iniFluxes = ini_fluxes
if ini_concs is not None:
iniConcs = read_initial_values(ini_concs, self.model.concids)
else:
iniConcs = ini_concs
optModel = InstMFAModel(self.model, fit_measured_fluxes, solver)
optModel.build_objective()
optModel.build_gradient()
optModel.build_flux_and_conc_bound_constraints()
optModel.build_initial_flux_and_conc_values(
ini_netfluxes = iniFluxes,
ini_concs = iniConcs
)
with Progress('INST fitting', silent = not show_progress):
res = optModel.solve_flux(tol, max_iters)
return InstFitResults(
*res[:8],
deepcopy(res[8]),
res[9],
deepcopy(res[10]),
*res[11:]
)
[docs]
def _solve_with_confidence_intervals(
self,
fit_measured_fluxes,
ini_fluxes,
ini_concs,
solver,
tol,
max_iters,
nruns
):
'''
Parameters
----------
fit_measured_fluxes: bool
Whether to fit measured fluxes.
ini_fluxes: ser or file in .tsv or .xlsx or None
Initial values of net fluxes.
ini_concs: ser or file in .tsv or .xlsx or None
Initial values of concentrations.
solvor: {"slsqp", "ralg"}
* If "slsqp", scipy.optimize.minimze will be used.
* If "ralg", openopt NLP solver will be used.
tol: float
Tolerance for termination.
max_iters: int
Maximum # of iterations.
nruns: int
# of estimations in each worker.
'''
import platform
if platform.system() == 'Linux':
import os
os.sched_setaffinity(os.getpid(), range(os.cpu_count()))
self._lambdify_matrix_As_and_Bs()
self._lambdify_matrix_Ms()
if ini_fluxes is not None:
iniFluxes = read_initial_values(ini_fluxes, self.model.netfluxids)
else:
iniFluxes = ini_fluxes
# if ini_concs is not None:
# iniConcs = read_initial_values(ini_concs, self.model.concids)
# else:
# iniConcs = ini_concs
optTotalfluxesSet = []
optNetfluxesSet = []
optConcsSet = []
for _ in range(nruns):
self.calculator._generate_random_fluxes()
self.calculator._generate_random_inst_MDVs()
optModel = InstMFAModel(self.model, fit_measured_fluxes, solver)
optModel.build_objective()
optModel.build_gradient()
optModel.build_flux_and_conc_bound_constraints()
optModel.build_initial_flux_and_conc_values(ini_netfluxes = iniFluxes)
try:
while True:
(optTotalfluxes,
optNetfluxes,
optConcs,
*_,
isSuccess
) = optModel.solve_flux(tol, max_iters)
if isSuccess:
break
except:
continue
optTotalfluxesSet.append(optTotalfluxes)
optNetfluxesSet.append(optNetfluxes)
optConcsSet.append(optConcs)
self.calculator._reset_measured_fluxes()
self.calculator._reset_measured_inst_MDVs()
return optTotalfluxesSet, optNetfluxesSet, optConcsSet
[docs]
def solve_with_confidence_intervals(
self,
fit_measured_fluxes = True,
ini_fluxes = None,
ini_concs = None,
solver = 'slsqp',
tol = 1e-6,
max_iters = 400,
n_runs = 100,
n_jobs = 1,
show_progress = True
):
'''
Parameters
----------
fit_measured_fluxes: bool
Whether to fit measured fluxes.
ini_fluxes: ser or file in .tsv or .xlsx
Initial values of net fluxes.
ini_concs: ser or file in .tsv or .xlsx
Initial values of concentrations.
solvor: {"slsqp", "ralg"}
* If "slsqp", scipy.optimize.minimze will be used.
* If "ralg", openopt NLP solver will be used.
tol: float
Tolerance for termination.
max_iters: int
Max # of iterations.
show_progress: bool
Whether to show the progress bar.
n_runs: int
# of runs to estimate confidence intervals.
n_jobs: int
# of jobs to run in parallel.
'''
self._check_dependencies(fit_measured_fluxes)
self._unset_matrix_As_and_Bs()
self._unset_matrix_Ms()
if n_runs <= n_jobs:
nruns_worker = 1
else:
nruns_worker = ceil(n_runs/n_jobs)
pool = Pool(processes = n_jobs)
with Progress('INST fitting with CIs', silent = not show_progress):
resSet = []
for _ in range(n_jobs):
res = pool.apply_async(
func = self._solve_with_confidence_intervals,
args = (
fit_measured_fluxes,
ini_fluxes,
ini_concs,
solver,
tol,
max_iters,
nruns_worker
)
)
resSet.append(res)
pool.close()
pool.join()
resSet = [res.get() for res in resSet]
totalFluxesSet = []
netFluxesSet = []
concsSet = []
for totalFluxesSubset, netFluxesSubset, concsSubset in resSet:
totalFluxesSet.extend(totalFluxesSubset)
netFluxesSet.extend(netFluxesSubset)
concsSet.extend(concsSubset)
self._lambdify_matrix_As_and_Bs()
self._lambdify_matrix_Ms()
return InstFitMCResults(totalFluxesSet, netFluxesSet, concsSet)