def policy_function(nn_model, state, v_log_counts_path, *args):
#last input argument should be epsilon for use when using greedy-epsilon algorithm.
varargin = args
nargin = len(varargin)
epsilon_greedy = 0.0
if (nargin == 1):
epsilon_greedy = varargin[0]
used_random_step = False
rxn_choices = [i for i in range(num_rxns)]
unif_rand = np.random.uniform(0, 1)
if ((unif_rand < epsilon_greedy) and (len(rxn_choices) > 0)):
used_random_step = True
random_choice = random.choice(rxn_choices)
final_action = random_choice
used_random_step = 1
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_path,
method='lm',
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs,
Keq_constant, state))
final_v_log_counts = res_lsq.x
new_log_metabolites = np.append(final_v_log_counts, f_log_counts)
final_state = state.copy()
newE = max_entropy_functions.calc_new_enzyme_simple(
state, final_action)
final_state = state.copy() #DO NOT MODIFY ORIGINAL STATE
final_state[final_action] = newE
final_delta_s_metab = max_entropy_functions.calc_deltaS_metab(
final_v_log_counts, target_v_log_counts)
final_KQ_f = max_entropy_functions.odds(
new_log_metabolites, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
final_KQ_r = max_entropy_functions.odds(
new_log_metabolites, mu0, -S_mat, P_mat, R_back_mat,
delta_increment_for_small_concs, Keq_inverse, -1)
value_current_state = state_value(
nn_model,
torch.from_numpy(final_state).float().to(device))
value_current_state = value_current_state.item()
final_reward = reward_value(final_v_log_counts, v_log_counts_path, \
final_KQ_f, final_KQ_r,\
final_state, state)
else:
#In this, we must choose base on the best prediction base on environmental feedback
v_log_counts = v_log_counts_path
log_metabolites = np.append(v_log_counts, f_log_counts)
rxn_flux = max_entropy_functions.oddsDiff(
v_log_counts, f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, state)
KQ_f = max_entropy_functions.odds(log_metabolites, mu0, S_mat,
R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r = max_entropy_functions.odds(log_metabolites, mu0, -S_mat, P_mat,
R_back_mat,
delta_increment_for_small_concs,
Keq_inverse, -1)
[RR, Jac
] = max_entropy_functions.calc_Jac2(v_log_counts, f_log_counts, S_mat,
delta_increment_for_small_concs,
KQ_f, KQ_r, state)
A = max_entropy_functions.calc_A(v_log_counts, f_log_counts, S_mat,
Jac, state)
delta_S_metab = max_entropy_functions.calc_deltaS_metab(
v_log_counts, target_v_log_counts)
[ccc, fcc] = max_entropy_functions.conc_flux_control_coeff(
nvar, A, S_mat, rxn_flux, RR)
indices = [i for i in range(0, len(Keq_constant))]
#minimal varialbes to run optimization
#variables=[state, v_log_counts, f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant]
#with Pool() as pool:
# async_result = pool.starmap(potential_step, zip(indices, repeat(variables)))
# pool.close()
# pool.join()
#end = time.time()
#print(v_log_counts_path)
async_result = pstep.dispatch(rxn_choices, S_mat, R_back_mat, P_mat,
Keq_constant, state, f_log_counts,
v_log_counts_path)
#print(async_result[0])
temp_action_value = -np.inf
for act in range(0, len(async_result)):
#new_v_log_counts = async_result[act][0] #output from poo
new_v_log_counts = async_result[act]
new_log_metabolites = np.append(new_v_log_counts, f_log_counts)
trial_state_sample = state.copy()
newE = max_entropy_functions.calc_new_enzyme_simple(state, act)
trial_state_sample = state.copy() #DO NOT MODIFY ORIGINAL STATE
trial_state_sample[act] = newE
new_delta_S_metab = max_entropy_functions.calc_deltaS_metab(
new_v_log_counts, target_v_log_counts)
KQ_f_new = max_entropy_functions.odds(
new_log_metabolites, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant)
KQ_r_new = max_entropy_functions.odds(
new_log_metabolites, mu0, -S_mat, P_mat, R_back_mat,
delta_increment_for_small_concs, Keq_inverse, -1)
value_current_state = state_value(
nn_model,
torch.from_numpy(trial_state_sample).float().to(device))
value_current_state = value_current_state.item()
current_reward = reward_value(new_v_log_counts, v_log_counts, \
KQ_f_new, KQ_r_new,\
trial_state_sample, state)
#print(current_reward)
action_value = current_reward + (gamma) * value_current_state
if (action_value > temp_action_value):
#then a new action is the best.
temp_action_value = action_value
#set best output variables
final_action = act
final_reward = current_reward
final_KQ_f = KQ_f_new
final_KQ_r = KQ_r_new
final_v_log_counts = new_v_log_counts
final_state = trial_state_sample
final_delta_s_metab = new_delta_S_metab
#print(final_state)
#print('final_delta_s_metab')
#print(final_delta_s_metab)
return [final_action,\
final_reward,\
final_KQ_f,\
final_KQ_r,\
final_v_log_counts,\
final_state,\
final_delta_s_metab,used_random_step,0.0,0.0]
delta_increment_for_small_concs,
Keq_constant, E_regulation)
KQ_f = max_entropy_functions.odds(log_metabolites, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs,
Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r = max_entropy_functions.odds(log_metabolites, mu0, -S_mat, P_mat,
R_back_mat,
delta_increment_for_small_concs,
Keq_inverse, -1)
epr = max_entropy_functions.entropy_production_rate(
KQ_f, KQ_r, E_regulation)
delta_S_metab = max_entropy_functions.calc_deltaS_metab(
v_log_counts, target_v_log_counts)
delta_S = max_entropy_functions.calc_deltaS(v_log_counts,
target_v_log_counts,
f_log_counts, S_mat, KQ_f)
[RR,
Jac] = max_entropy_functions.calc_Jac2(v_log_counts, f_log_counts, S_mat,
delta_increment_for_small_concs,
KQ_f, KQ_r, E_regulation)
A = max_entropy_functions.calc_A(v_log_counts, f_log_counts, S_mat, Jac,
E_regulation)
[ccc, fcc
] = max_entropy_functions.conc_flux_control_coeff(nvar, A, S_mat,
rxn_flux, RR)
def sarsa_n(nn_model, loss_fn, optimizer, scheduler, state_sample, n_back_step,
epsilon_greedy):
total_time_cpu = 0
total_time_nn = 0
#reset for each episode. policy will add
random_steps_taken = 0
nn_steps_taken = 0
final_state = []
final_KQ_f = []
final_KQ_r = []
reached_terminal_state = False
average_loss = []
final_reward = 0
sum_reward_episode = 0
end_of_path = 5000 #this is the maximum length a path can take
KQ_f_matrix = np.zeros(shape=(num_rxns, end_of_path + 1))
KQ_r_matrix = np.zeros(shape=(num_rxns, end_of_path + 1))
states_matrix = np.zeros(shape=(num_rxns, end_of_path + 1))
delta_S_metab_matrix = np.zeros(shape=(nvar, end_of_path + 1))
v_log_counts_matrix = np.zeros(shape=(nvar, end_of_path + 1))
states_matrix[:, 0] = state_sample
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_static,
method='lm',
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, states_matrix[:, 0]))
v_log_counts_matrix[:, 0] = res_lsq.x.copy()
log_metabolites = np.append(v_log_counts_matrix[:, 0], f_log_counts)
rxn_flux_init = max_entropy_functions.oddsDiff(
v_log_counts_matrix[:, 0], f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, states_matrix[:, 0])
KQ_f_matrix[:, 0] = max_entropy_functions.odds(
log_metabolites, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r_matrix[:, 0] = max_entropy_functions.odds(
log_metabolites, mu0, -S_mat, P_mat, R_back_mat,
delta_increment_for_small_concs, Keq_inverse, -1)
delta_S_metab_matrix[:, 0] = max_entropy_functions.calc_deltaS_metab(
v_log_counts_matrix[:, 0], target_v_log_counts)
reward_vec = np.zeros(end_of_path + 1)
reward_vec[0] = 0.0
rxn_flux_path = rxn_flux_init.copy()
for t in range(0, end_of_path):
if (t < end_of_path):
#This represents the choice from the current policy.
[React_Choice,reward_vec[t+1],\
KQ_f_matrix[:,t+1], KQ_r_matrix[:,t+1],\
v_log_counts_matrix[:,t+1],\
states_matrix[:,t+1],\
delta_S_metab_matrix[:,t+1],\
used_random_step,time_cpu,time_nn] = policy_function(nn_model, states_matrix[:,t], v_log_counts_matrix[:,t], epsilon_greedy)#regulate each reaction.
total_time_cpu += time_cpu
total_time_nn += time_nn
if (used_random_step):
random_steps_taken += 1
else:
nn_steps_taken += 1
if (React_Choice == -1):
print("bad reaction choice, using action = -1")
break
rxn_flux_path = max_entropy_functions.oddsDiff(
v_log_counts_matrix[:, t + 1], f_log_counts, mu0, S_mat,
R_back_mat, P_mat, delta_increment_for_small_concs,
Keq_constant, states_matrix[:, t + 1])
if (np.max(rxn_flux_path) < 1.0):
print("draining flux")
break
epr_path = max_entropy_functions.entropy_production_rate(
KQ_f_matrix[:, t + 1], KQ_r_matrix[:, t + 1],
states_matrix[:, t + 1])
sum_reward_episode += reward_vec[t + 1]
current_state = states_matrix[:, t + 1].copy()
#We stop the path if we have no more positive loss function values, or if we revisit a state.
if ((delta_S_metab_matrix[:, t + 1] <= 0.0).all()):
end_of_path = t + 1 #stops simulation at step t+1
reached_terminal_state = True
final_state = states_matrix[:, t + 1].copy()
final_KQ_f = KQ_f_matrix[:, t + 1].copy()
final_KQ_r = KQ_r_matrix[:, t + 1].copy()
final_reward = epr_path
print(
"**************************************Path Length ds<0******************************************"
)
print(end_of_path)
print("Final STATE")
print(states_matrix[:, t + 1])
print(rxn_flux_path)
print("original epr")
print(epr_path)
print("all rewards")
print(reward_vec[0:t + 1])
##BEGIN LEARNING
tau = t - n_back_step + 1
if (tau >= 0):
#breakpoint()
estimate_value = torch.zeros(1, device=device)
for i in range(tau + 1, min(tau + n_back_step, end_of_path) + 1):
estimate_value += (gamma**(i - tau - 1)) * reward_vec[i]
if ((tau + n_back_step) < end_of_path):
begin_nn = time.time()
value_tau_n = state_value(
nn_model,
torch.from_numpy(
states_matrix[:,
tau + n_back_step]).float().to(device))
end_nn = time.time()
total_time_nn += end_nn - begin_nn
estimate_value += (gamma**(n_back_step)) * value_tau_n
begin_nn = time.time()
value_tau = state_value(
nn_model,
torch.from_numpy(states_matrix[:, tau]).float().to(device))
end_nn = time.time()
total_time_nn += end_nn - begin_nn
if (value_tau.requires_grad == False):
breakpoint()
if (estimate_value.requires_grad == True):
estimate_value.detach_()
#WARNING
#loss ordering should be input with requires_grad == True,
#followed by target with requires_grad == False
#breakpoint()
begin_nn = time.time()
loss = loss_fn(value_tau, estimate_value) #MSE
optimizer.zero_grad()
loss.backward()
clipping_value = 1.0
torch.nn.utils.clip_grad_norm_(nn_model.parameters(),
clipping_value)
optimizer.step()
end_nn = time.time()
total_time_nn += end_nn - begin_nn
average_loss.append(loss.item())
if (tau >= (end_of_path - 1)):
break
#after episode is finished, take average loss
average_loss_episode = np.mean(average_loss)
print("index of max error on path")
print(average_loss.index(max(average_loss)))
return [sum_reward_episode, average_loss_episode,max(average_loss),final_reward, final_state, final_KQ_f,final_KQ_r,\
reached_terminal_state, random_steps_taken,nn_steps_taken]
def run(argv):
try:
os.makedirs(cwd + '/GLYCOLYSIS_TCA_GOGAT/data')
except FileExistsError:
# directory already exists
pass
try:
os.makedirs(cwd + '/GLYCOLYSIS_TCA_GOGAT/models_final_data')
except FileExistsError:
# directory already exists
pass
pd.set_option('display.max_columns', None, 'display.max_rows', None)
###Default Values
#If no experimental data is available, we can estimate using 'rule-of-thumb' data at 0.001
use_experimental_data = False
learning_rate = 1e-8 #3rd
epsilon = 0.5 #4th
eps_threshold = 25 #5th
gamma = 0.9 #6th
updates = 500
penalty_reward_scalar = 0.0
#load input
total = len(sys.argv)
cmdargs = str(sys.argv)
print("The total numbers of args passed to the script: %d " % total)
print("Args list: %s " % cmdargs)
print("Script name: %s" % str(sys.argv[0]))
for i in range(total):
print("Argument # %d : %s" % (i, str(sys.argv[i])))
sim_number = int(sys.argv[1])
n_back_step = int(sys.argv[2])
if (n_back_step < 1):
print('n must be larger than zero')
return
if (total > 3):
use_experimental_data = bool(int(sys.argv[3]))
if (total > 4):
learning_rate = float(sys.argv[4])
if (total > 5):
epsilon = float(sys.argv[5])
if (total > 6):
eps_threshold = float(sys.argv[6])
if (total > 7):
gamma = float(sys.argv[7])
print("sim")
print(sim_number)
print("n_back_step")
print(n_back_step)
print("using experimental metabolite data")
print(use_experimental_data)
print("learning_rate")
print(learning_rate)
print("epsilon")
print(epsilon)
print("eps_threshold")
print(eps_threshold)
print("gamma")
print(gamma)
#Initial Values
T = 298.15
R = 8.314e-03
RT = R * T
N_avogadro = 6.022140857e+23
VolCell = 1.0e-15
Concentration2Count = N_avogadro * VolCell
concentration_increment = 1 / (N_avogadro * VolCell)
np.set_printoptions(suppress=True) #turn off printin
fdat = open(cwd + '/GLYCOLYSIS_TCA_GOGAT/GLYCOLYSIS_TCA_GOGAT.dat', 'r')
left = 'LEFT'
right = 'RIGHT'
left_compartment = 'LEFT_COMPARTMENT'
right_compartment = 'RIGHT_COMPARTMENT'
enzyme_level = 'ENZYME_LEVEL'
deltag0 = 'DGZERO'
deltag0_sigma = 'DGZERO StdDev'
same_compartment = 'Same Compartment?'
full_rxn = 'Full Rxn'
reactions = pd.DataFrame(index=[],
columns=[
left, right, left_compartment,
right_compartment, enzyme_level, deltag0,
deltag0_sigma, same_compartment, full_rxn
])
reactions.index.name = 'REACTION'
S_matrix = pd.DataFrame(index=[], columns=[enzyme_level])
S_matrix.index.name = 'REACTION'
for line in fdat:
if (line.startswith('REACTION')):
rxn_name = line[9:-1].lstrip()
S_matrix.loc[rxn_name, enzyme_level] = 1.0
reactions.loc[rxn_name, enzyme_level] = 1.0
if (re.match("^LEFT\s", line)):
line = line.upper()
left_rxn = line[4:-1].lstrip()
left_rxn = re.sub(r'\s+$', '',
left_rxn) #Remove trailing white space
reactions.loc[rxn_name, left] = left_rxn
elif (re.match('^RIGHT\s', line)):
line = line.upper()
right_rxn = line[5:-1].lstrip()
right_rxn = re.sub(r'\s+$', '',
right_rxn) #Remove trailing white space
reactions.loc[rxn_name, right] = right_rxn
elif (line.startswith(left_compartment)):
cpt_name = line[16:-1].lstrip()
reactions.loc[rxn_name, left_compartment] = cpt_name
reactants = re.split(' \+ ', left_rxn)
for idx in reactants:
values = re.split(' ', idx)
if len(values) == 2:
stoichiometry = np.float64(values[0])
molecule = values[1]
if not re.search(':', molecule):
molecule = molecule + ':' + cpt_name
else:
stoichiometry = np.float64(-1.0)
molecule = values[0]
if not re.search(':', molecule):
molecule = molecule + ':' + cpt_name
S_matrix.loc[rxn_name, molecule] = stoichiometry
elif (line.startswith(right_compartment)):
cpt_name = line[17:-1].lstrip()
reactions.loc[rxn_name, right_compartment] = cpt_name
products = re.split(' \+ ', right_rxn)
for idx in products:
values = re.split(' ', idx)
if len(values) == 2:
stoichiometry = np.float64(values[0])
molecule = values[1]
if not re.search(':', molecule):
molecule = molecule + ':' + cpt_name
else:
stoichiometry = np.float64(1.0)
molecule = values[0]
if not re.search(':', molecule):
molecule = molecule + ':' + cpt_name
S_matrix.loc[rxn_name, molecule] = stoichiometry
elif (re.match("^ENZYME_LEVEL\s", line)):
level = line[12:-1].lstrip()
reactions.loc[rxn_name, enzyme_level] = float(level)
S_matrix.loc[rxn_name, enzyme_level] = float(level)
elif re.match('^COMMENT', line):
continue
elif re.match(r'//', line):
continue
elif re.match('^#', line):
continue
fdat.close()
S_matrix.fillna(0, inplace=True)
S_active = S_matrix[S_matrix[enzyme_level] > 0.0]
active_reactions = reactions[reactions[enzyme_level] > 0.0]
del S_active[enzyme_level]
S_active = S_active.loc[:, (S_active != 0).any(axis=0)]
np.shape(S_active.values)
reactions[full_rxn] = reactions[left] + ' = ' + reactions[right]
if (1):
for idx in reactions.index:
boltzmann_rxn_str = reactions.loc[idx, 'Full Rxn']
if re.search(':', boltzmann_rxn_str):
all_cmprts = re.findall(':\S+', boltzmann_rxn_str)
[s.replace(':', '') for s in all_cmprts] # remove all the ':'s
different_compartments = 0
for cmpt in all_cmprts:
if not re.match(all_cmprts[0], cmpt):
different_compartments = 1
if ((not different_compartments)
and (reactions[left_compartment].isnull
or reactions[right_compartment].isnull)):
reactions.loc[idx, left_compartment] = cmpt
reactions.loc[idx, right_compartment] = cmpt
reactions.loc[idx, same_compartment] = True
if different_compartments:
reactions.loc[idx, same_compartment] = False
else:
if (reactions.loc[idx, left_compartment] == reactions.loc[
idx, right_compartment]):
reactions.loc[idx, same_compartment] = True
else:
reactions.loc[idx, same_compartment] = False
# ## Calculate Standard Free Energies of Reaction
reactions.loc['CSm', deltag0] = -35.1166
reactions.loc['ACONTm', deltag0] = 7.62949
reactions.loc['ICDHxm', deltag0] = -2.872
reactions.loc['AKGDam', deltag0] = -36.3549
reactions.loc['SUCOASm', deltag0] = 1.924481
reactions.loc['SUCD1m', deltag0] = 0
reactions.loc['FUMm', deltag0] = -3.44873
reactions.loc['MDHm', deltag0] = 29.9942
reactions.loc['GAPD', deltag0] = 6.68673
reactions.loc['PGK', deltag0] = -18.4733
reactions.loc['TPI', deltag0] = 5.48642
reactions.loc['FBA', deltag0] = 20.5096
reactions.loc['PYK', deltag0] = -27.5366
reactions.loc['PGM', deltag0] = 4.19953
reactions.loc['ENO', deltag0] = -4.08222
reactions.loc['HEX1', deltag0] = -17.0578
reactions.loc['PGI', deltag0] = 2.52401
reactions.loc['PFK', deltag0] = -15.4549
reactions.loc['PYRt2m', deltag0] = -RT * np.log(10)
reactions.loc['PDHm', deltag0] = -43.9219
reactions.loc['GOGAT', deltag0] = 48.8552
reactions.loc['CSm', deltag0_sigma] = 0.930552
reactions.loc['ACONTm', deltag0_sigma] = 0.733847
reactions.loc['ICDHxm', deltag0_sigma] = 7.62095
reactions.loc['AKGDam', deltag0_sigma] = 7.97121
reactions.loc['SUCOASm', deltag0_sigma] = 1.48197
reactions.loc['SUCD1m', deltag0_sigma] = 2.31948
reactions.loc['FUMm', deltag0_sigma] = 0.607693
reactions.loc['MDHm', deltag0_sigma] = 0.422376
reactions.loc['GAPD', deltag0_sigma] = 0.895659
reactions.loc['PGK', deltag0_sigma] = 0.889982
reactions.loc['TPI', deltag0_sigma] = 0.753116
reactions.loc['FBA', deltag0_sigma] = 0.87227
reactions.loc['PYK', deltag0_sigma] = 0.939774
reactions.loc['PGM', deltag0_sigma] = 0.65542
reactions.loc['ENO', deltag0_sigma] = 0.734193
reactions.loc['HEX1', deltag0_sigma] = 0.715237
reactions.loc['PGI', deltag0_sigma] = 0.596775
reactions.loc['PFK', deltag0_sigma] = 0.886629
reactions.loc['PYRt2m', deltag0_sigma] = 0
reactions.loc['PDHm', deltag0_sigma] = 7.66459
reactions.loc['GOGAT', deltag0_sigma] = 2.0508
# ## Set Fixed Concentrations/Boundary Conditions
conc = 'Conc'
variable = 'Variable'
conc_exp = 'Conc_Experimental'
metabolites = pd.DataFrame(index=S_active.columns,
columns=[conc, conc_exp, variable])
metabolites[conc] = 0.001
metabolites[variable] = True
# Set the fixed metabolites:
metabolites.loc['ATP:MITOCHONDRIA', conc] = 9.600000e-03
metabolites.loc['ATP:MITOCHONDRIA', variable] = False
metabolites.loc['ADP:MITOCHONDRIA', conc] = 5.600000e-04
metabolites.loc['ADP:MITOCHONDRIA', variable] = False
metabolites.loc['ORTHOPHOSPHATE:MITOCHONDRIA', conc] = 2.000000e-02
metabolites.loc['ORTHOPHOSPHATE:MITOCHONDRIA', variable] = False
metabolites.loc['ATP:CYTOSOL', conc] = 9.600000e-03
metabolites.loc['ATP:CYTOSOL', variable] = False
metabolites.loc['ADP:CYTOSOL', conc] = 5.600000e-04
metabolites.loc['ADP:CYTOSOL', variable] = False
metabolites.loc['ORTHOPHOSPHATE:CYTOSOL', conc] = 2.000000e-02
metabolites.loc['ORTHOPHOSPHATE:CYTOSOL', variable] = False
metabolites.loc['NADH:MITOCHONDRIA', conc] = 8.300000e-05
metabolites.loc['NADH:MITOCHONDRIA', variable] = False
metabolites.loc['NAD+:MITOCHONDRIA', conc] = 2.600000e-03
metabolites.loc['NAD+:MITOCHONDRIA', variable] = False
metabolites.loc['NADH:CYTOSOL', conc] = 8.300000e-05
metabolites.loc['NADH:CYTOSOL', variable] = False
metabolites.loc['NAD+:CYTOSOL', conc] = 2.600000e-03
metabolites.loc['NAD+:CYTOSOL', variable] = False
metabolites.loc['ACETYL-COA:MITOCHONDRIA', conc] = 6.06E-04
metabolites.loc['ACETYL-COA:MITOCHONDRIA', variable] = True
metabolites.loc['COA:MITOCHONDRIA', conc] = 1.400000e-03
metabolites.loc['COA:MITOCHONDRIA', variable] = False
metabolites.loc['CO2:MITOCHONDRIA', conc] = 1.000000e-04
metabolites.loc['CO2:MITOCHONDRIA', variable] = False
metabolites.loc['H2O:MITOCHONDRIA', conc] = 55.5
metabolites.loc['H2O:MITOCHONDRIA', variable] = False
metabolites.loc['H2O:CYTOSOL', conc] = 55.5
metabolites.loc['H2O:CYTOSOL', variable] = False
metabolites.loc['BETA-D-GLUCOSE:CYTOSOL', conc] = 2.000000e-03
metabolites.loc['BETA-D-GLUCOSE:CYTOSOL', variable] = False
metabolites.loc['L-GLUTAMATE:MITOCHONDRIA', conc] = 9.60e-05
metabolites.loc['L-GLUTAMATE:MITOCHONDRIA', variable] = False
metabolites.loc['L-GLUTAMINE:MITOCHONDRIA', conc] = 3.81e-03
metabolites.loc['L-GLUTAMINE:MITOCHONDRIA', variable] = False
#When loading experimental concentrations, first copy current
#rule of thumb then overwrite with data values.
metabolites[conc_exp] = metabolites[conc]
metabolites.loc['(S)-MALATE:MITOCHONDRIA', conc_exp] = 1.68e-03
metabolites.loc['BETA-D-GLUCOSE-6-PHOSPHATE:CYTOSOL', conc_exp] = 7.88e-03
metabolites.loc['D-GLYCERALDEHYDE-3-PHOSPHATE:CYTOSOL',
conc_exp] = 2.71e-04
metabolites.loc['PYRUVATE:MITOCHONDRIA', conc_exp] = 3.66e-03
metabolites.loc['ISOCITRATE:MITOCHONDRIA', conc_exp] = 1.000000e-03
metabolites.loc['OXALOACETATE:MITOCHONDRIA', conc_exp] = 1.000000e-03
metabolites.loc['3-PHOSPHO-D-GLYCEROYL_PHOSPHATE:CYTOSOL',
conc_exp] = 1.000000e-03
metabolites.loc['ACETYL-COA:MITOCHONDRIA', conc_exp] = 6.06e-04
metabolites.loc['CITRATE:MITOCHONDRIA', conc_exp] = 1.96e-03
metabolites.loc['2-OXOGLUTARATE:MITOCHONDRIA', conc_exp] = 4.43e-04
metabolites.loc['FUMARATE:MITOCHONDRIA', conc_exp] = 1.15e-04
metabolites.loc['SUCCINYL-COA:MITOCHONDRIA', conc_exp] = 2.33e-04
metabolites.loc['3-PHOSPHO-D-GLYCERATE:CYTOSOL', conc_exp] = 1.54e-03
metabolites.loc['GLYCERONE_PHOSPHATE:CYTOSOL', conc_exp] = 3.060000e-03
metabolites.loc['SUCCINATE:MITOCHONDRIA', conc_exp] = 5.69e-04
metabolites.loc['PHOSPHOENOLPYRUVATE:CYTOSOL', conc_exp] = 1.84e-04
metabolites.loc['D-FRUCTOSE_1,6-BISPHOSPHATE:CYTOSOL', conc_exp] = 1.52e-02
metabolites.loc['D-FRUCTOSE_6-PHOSPHATE:CYTOSOL', conc_exp] = 2.52e-03
metabolites.loc['PYRUVATE:CYTOSOL', conc_exp] = 3.66E-03
metabolites.loc['2-PHOSPHO-D-GLYCERATE:CYTOSOL', conc_exp] = 9.180e-05
#%%
nvariables = metabolites[metabolites[variable]].count()
nvar = nvariables[variable]
metabolites.sort_values(
by=variable,
axis=0,
ascending=False,
inplace=True,
)
# ## Prepare model for optimization
# - Adjust S Matrix to use only reactions with activity > 0, if necessary.
# - Water stoichiometry in the stiochiometric matrix needs to be set to zero since water is held constant.
# - The initial concentrations of the variable metabolites are random.
# - All concentrations are changed to log counts.
# - Equilibrium constants are calculated from standard free energies of reaction.
# - R (reactant) and P (product) matrices are derived from S.
# Make sure all the indices and columns are in the correct order:
active_reactions = reactions[reactions[enzyme_level] > 0.0]
Sactive_index = S_active.index
active_reactions.reindex(index=Sactive_index, copy=False)
S_active = S_active.reindex(columns=metabolites.index, copy=False)
S_active['H2O:MITOCHONDRIA'] = 0
S_active['H2O:CYTOSOL'] = 0
where_are_NaNs = np.isnan(S_active)
S_active[where_are_NaNs] = 0
S_mat = S_active.values
Keq_constant = np.exp(-active_reactions[deltag0].astype('float') / RT)
Keq_constant = Keq_constant.values
P_mat = np.where(S_mat > 0, S_mat, 0)
R_back_mat = np.where(S_mat < 0, S_mat, 0)
E_regulation = np.ones(
Keq_constant.size
) # THis is the vector of enzyme activities, Range: 0 to 1.
mu0 = 1 #Dummy parameter for now; reserved for free energies of formation
conc_type = conc
if (use_experimental_data):
print("USING EXPERIMENTAL DATA")
conc_type = conc_exp
variable_concs = np.array(metabolites[conc_type].iloc[0:nvar].values,
dtype=np.float64)
v_log_concs = -10 + 10 * np.random.rand(
nvar) #Vary between 1 M to 1.0e-10 M
v_concs = np.exp(v_log_concs)
v_log_counts_stationary = np.log(v_concs * Concentration2Count)
v_log_counts = v_log_counts_stationary
#display(v_log_counts)
fixed_concs = np.array(metabolites[conc_type].iloc[nvar:].values,
dtype=np.float64)
fixed_counts = fixed_concs * Concentration2Count
f_log_counts = np.log(fixed_counts)
complete_target_log_counts = np.log(Concentration2Count *
metabolites[conc_type].values)
target_v_log_counts = complete_target_log_counts[0:nvar]
target_f_log_counts = complete_target_log_counts[nvar:]
delta_increment_for_small_concs = (10**-50) * np.zeros(
metabolites[conc_type].values.size)
variable_concs_begin = np.array(metabolites[conc_type].iloc[0:nvar].values,
dtype=np.float64)
#%% Basic test
v_log_counts = np.log(variable_concs_begin * Concentration2Count)
E_regulation = np.ones(
Keq_constant.size
) # THis is the vector of enzyme activities, Range: 0 to 1.
nvar = v_log_counts.size
#WARNING: INPUT LOG_COUNTS TO ALL FUNCTIONS. CONVERSION TO COUNTS IS DONE INTERNALLY
res_lsq1 = least_squares(max_entropy_functions.derivatives,
v_log_counts,
method='lm',
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, E_regulation))
res_lsq2 = least_squares(max_entropy_functions.derivatives,
v_log_counts,
method='dogbox',
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, E_regulation))
rxn_flux = max_entropy_functions.oddsDiff(res_lsq1.x, f_log_counts, mu0,
S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, E_regulation)
begin_log_metabolites = np.append(res_lsq1.x, f_log_counts)
E_regulation = np.ones(
Keq_constant.size
) # THis is the vector of enzyme activities, Range: 0 to 1.
log_metabolites = np.append(res_lsq1.x, f_log_counts)
KQ_f = max_entropy_functions.odds(log_metabolites, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs,
Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r = max_entropy_functions.odds(log_metabolites, mu0, -S_mat, P_mat,
R_back_mat,
delta_increment_for_small_concs,
Keq_inverse, -1)
[RR,
Jac] = max_entropy_functions.calc_Jac2(res_lsq1.x, f_log_counts, S_mat,
delta_increment_for_small_concs,
KQ_f, KQ_r, E_regulation)
A = max_entropy_functions.calc_A(res_lsq1.x, f_log_counts, S_mat, Jac,
E_regulation)
[ccc, fcc
] = max_entropy_functions.conc_flux_control_coeff(nvar, A, S_mat,
rxn_flux, RR)
React_Choice = 6
newE = max_entropy_functions.calc_reg_E_step(E_regulation, React_Choice,
nvar, res_lsq1.x,
f_log_counts,
complete_target_log_counts,
S_mat, A, rxn_flux, KQ_f)
delta_S_metab = max_entropy_functions.calc_deltaS_metab(
res_lsq1.x, target_v_log_counts)
ipolicy = 7 #use ipolicy=1 or 4
#%% END Basic test
#Machine learning
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print('Using device:', device)
#set variables in ML program
me.cwd = cwd
me.device = device
me.v_log_counts_static = v_log_counts_stationary
me.target_v_log_counts = target_v_log_counts
me.complete_target_log_counts = complete_target_log_counts
me.Keq_constant = Keq_constant
me.f_log_counts = f_log_counts
me.P_mat = P_mat
me.R_back_mat = R_back_mat
me.S_mat = S_mat
me.delta_increment_for_small_concs = delta_increment_for_small_concs
me.nvar = nvar
me.mu0 = mu0
me.gamma = gamma
me.num_rxns = Keq_constant.size
me.penalty_reward_scalar = penalty_reward_scalar
#%%
N, D_in, H, D_out = 1, Keq_constant.size, 20 * Keq_constant.size, 1
#create neural network
nn_model = torch.nn.Sequential(torch.nn.Linear(D_in, H), torch.nn.Tanh(),
torch.nn.Linear(H, D_out)).to(device)
loss_fn = torch.nn.MSELoss(reduction='sum')
#loss_fn = torch.nn.L1Loss()
#learning_rate=5e-6
#optimizer = torch.optim.SGD(nn_model.parameters(), lr=learning_rate, momentum=0.9)
optimizer = torch.optim.SGD(nn_model.parameters(),
lr=learning_rate,
momentum=0.9)
#optimizer = torch.optim.Adam(nn_model.parameters(), lr=3e-4)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer,
patience=500,
verbose=True,
min_lr=1e-10,
cooldown=10,
threshold=1e-5)
v_log_counts = v_log_counts_stationary.copy()
episodic_loss = []
episodic_loss_max = []
episodic_epr = []
episodic_reward = []
episodic_nn_step = []
episodic_random_step = []
epsilon_greedy_init = epsilon
final_states = np.zeros(Keq_constant.size)
final_KQ_fs = np.zeros(Keq_constant.size)
final_KQ_rs = np.zeros(Keq_constant.size)
epr_per_state = []
for update in range(0, updates):
x_changing = 1 * torch.rand(1000, D_in, device=device)
#generate state to use
state_sample = np.zeros(Keq_constant.size)
for sample in range(0, len(state_sample)):
state_sample[sample] = np.random.uniform(1, 1)
#annealing test
if ((update % eps_threshold == 0) and (update != 0)):
epsilon = epsilon / 2
print("RESET epsilon ANNEALING")
print(epsilon)
prediction_x_changing_previous = nn_model(x_changing)
#nn_model.train()
[sum_reward, average_loss,max_loss,final_epr,final_state,final_KQ_f,final_KQ_r, reached_terminal_state,\
random_steps_taken,nn_steps_taken] = me.sarsa_n(nn_model,loss_fn, optimizer, scheduler, state_sample, n_back_step, epsilon)
print("EPISODE")
print(update)
print("MAXIMUM LAYER WEIGHTS")
for layer in nn_model.modules():
try:
print(torch.max(layer.weight))
except:
print("")
print('random,nn steps')
print(random_steps_taken)
print(nn_steps_taken)
if (reached_terminal_state):
final_states = np.vstack((final_states, final_state))
final_KQ_fs = np.vstack((final_KQ_fs, final_KQ_f))
final_KQ_rs = np.vstack((final_KQ_rs, final_KQ_r))
epr_per_state.append(final_epr)
episodic_epr.append(final_epr)
episodic_loss.append(average_loss)
episodic_loss_max.append(max_loss)
episodic_reward.append(sum_reward)
episodic_nn_step.append(nn_steps_taken)
episodic_random_step.append(random_steps_taken)
scheduler.step(average_loss)
print("TOTAL REWARD")
print(sum_reward)
print("ave loss")
print(average_loss)
print("max_loss")
print(max_loss)
print(optimizer.state_dict)
print(scheduler.state_dict())
prediction_x_changing = nn_model(x_changing)
total_prediction_changing_diff = sum(
abs(prediction_x_changing - prediction_x_changing_previous))
print("TOTALPREDICTION")
print(total_prediction_changing_diff)
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/data/' +
'temp_episodic_loss_' + str(n_back_step) + '_lr' +
str(learning_rate) + '_' + str(eps_threshold) + '_eps' +
str(epsilon_greedy_init) + '_' + str(sim_number) +
'_penalty_reward_scalar_' + str(me.penalty_reward_scalar) +
'_use_experimental_metab_' +
str(int(use_experimental_data)) + '.txt',
episodic_loss,
fmt='%f')
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/data/' + 'temp_epr_' +
str(n_back_step) + '_lr' + str(learning_rate) + '_' +
str(eps_threshold) + '_eps' + str(epsilon_greedy_init) +
'_' + str(sim_number) + '_penalty_reward_scalar_' +
str(me.penalty_reward_scalar) + '_use_experimental_metab_' +
str(int(use_experimental_data)) + '.txt',
episodic_epr,
fmt='%f')
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/data/' +
'temp_episodic_reward_' + str(n_back_step) + '_lr' +
str(learning_rate) + '_' + str(eps_threshold) + '_eps' +
str(epsilon_greedy_init) + '_' + str(sim_number) +
'_penalty_reward_scalar_' + str(me.penalty_reward_scalar) +
'_use_experimental_metab_' +
str(int(use_experimental_data)) + '.txt',
episodic_reward,
fmt='%f')
if (update > 200):
if ((max(episodic_loss[-100:]) - min(episodic_loss[-100:]) < 0.025)
and (update > 350)):
break
#%%
#gamma9 -> gamma=0.9
#n8 -> n_back_step=8
#k5 -> E=E-E/5 was used
#lr5e6 -> begin lr=0.5*e-6
torch.save(nn_model, cwd+'/GLYCOLYSIS_TCA_GOGAT/models_final_data/'+
'complete_model_gly_tca_gog_gamma9_n'+str(n_back_step)+'_k5_'\
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_' +str(int(use_experimental_data))+
'_sim'+str(sim_number) + '.pth')
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/models_final_data/' +
'episodic_loss_gamma9_n' + str(n_back_step) + '_k5_'
'_lr' + str(learning_rate) + '_threshold' + str(eps_threshold) +
'_eps' + str(epsilon_greedy_init) + '_penalty_reward_scalar_' +
str(me.penalty_reward_scalar) + '_use_experimental_metab_' +
str(int(use_experimental_data)) + '_sim' + str(sim_number) +
'.txt',
episodic_loss,
fmt='%f')
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/models_final_data/' +
'episodic_loss_max_gamma9_n' + str(n_back_step) + '_k5_' +
'_lr' + str(learning_rate) + '_threshold' + str(eps_threshold) +
'_eps' + str(epsilon_greedy_init) + '_penalty_reward_scalar_' +
str(me.penalty_reward_scalar) + '_use_experimental_metab_' +
str(int(use_experimental_data)) + '_sim' + str(sim_number) +
'.txt',
episodic_loss_max,
fmt='%f')
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/models_final_data/' +
'episodic_reward_gamma9_n' + str(n_back_step) + '_k5_' + '_lr' +
str(learning_rate) + '_threshold' + str(eps_threshold) +
'_eps' + str(epsilon_greedy_init) + '_penalty_reward_scalar_' +
str(me.penalty_reward_scalar) + '_use_experimental_metab_' +
str(int(use_experimental_data)) + '_sim' + str(sim_number) +
'.txt',
episodic_reward,
fmt='%f')
np.savetxt(cwd+'/GLYCOLYSIS_TCA_GOGAT/models_final_data/'+
'final_states_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+\
'.txt', final_states, fmt='%f')
np.savetxt(cwd+'/GLYCOLYSIS_TCA_GOGAT/models_final_data/'+
'final_KQF_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+\
'.txt', final_KQ_fs, fmt='%f')
np.savetxt(cwd+'/GLYCOLYSIS_TCA_GOGAT/models_final_data/'+
'final_KQR_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+\
'.txt', final_KQ_rs, fmt='%f')
np.savetxt(cwd + '/GLYCOLYSIS_TCA_GOGAT/models_final_data/' +
'epr_per_state_gamma9_n' + str(n_back_step) + '_k5_' + '_lr' +
str(learning_rate) + '_threshold' + str(eps_threshold) +
'_eps' + str(epsilon_greedy_init) + '_penalty_reward_scalar_' +
str(me.penalty_reward_scalar) + '_use_experimental_metab_' +
str(int(use_experimental_data)) + '_sim' + str(sim_number) +
'.txt',
epr_per_state,
fmt='%f')
Keq_inverse = np.power(Keq_constant,-1)
KQ_r = max_entropy_functions.odds(log_metabolites,mu0,-S_mat, P_mat, R_back_mat, delta_increment_for_small_concs,Keq_inverse,-1)
[RR,Jac] = max_entropy_functions.calc_Jac2(res_lsq1.x, f_log_counts, S_mat, delta_increment_for_small_concs, KQ_f, KQ_r, E_regulation)
A = max_entropy_functions.calc_A(res_lsq1.x,f_log_counts, S_mat, Jac, E_regulation )
[ccc,fcc] = max_entropy_functions.conc_flux_control_coeff(nvar, A, S_mat, rxn_flux, RR)
React_Choice=6
newE = max_entropy_functions.calc_reg_E_step(E_regulation,React_Choice, nvar, res_lsq1.x, f_log_counts, complete_target_log_counts,
S_mat, A, rxn_flux,KQ_f)
delta_S_metab = max_entropy_functions.calc_deltaS_metab(res_lsq1.x, target_v_log_counts);
ipolicy = 7 #use ipolicy=1 or 4
reaction_choice = max_entropy_functions.get_enzyme2regulate(ipolicy, delta_S_metab, ccc, KQ_f, E_regulation, res_lsq1.x)
display(newE)
display(reaction_choice)
#%%
import torch
device = torch.device("cpu")
import machine_learning_functions as me
gamma = 0.9
def policy_function(nn_model, state, v_log_counts_path, *args):
#last input argument should be epsilon for use when using greedy-epsilon algorithm.
KQ_f_matrix = np.zeros(shape=(num_rxns, num_rxns))
KQ_r_matrix = np.zeros(shape=(num_rxns, num_rxns))
states_matrix = np.zeros(shape=(num_rxns, num_rxns))
delta_S_metab_matrix = np.zeros(shape=(nvar, num_rxns))
v_log_counts_matrix = np.zeros(shape=(nvar, num_rxns))
varargin = args
nargin = len(varargin)
epsilon_greedy = 0.0
if (nargin == 1):
epsilon_greedy = varargin[0]
rxn_choices = [i for i in range(num_rxns)]
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_path,
method=Method1,
bounds=(-500, 500),
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, state))
if (res_lsq.optimality > 1e-05):
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_path,
method=Method2,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs,
Keq_constant, state))
if (res_lsq.optimality > 1e-05):
res_lsq = least_squares(
max_entropy_functions.derivatives,
v_log_counts_path,
method=Method3,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, state))
#v_log_counts = v_log_counts_path.copy()
v_log_counts = res_lsq.x
if (np.sum(np.abs(v_log_counts - v_log_counts_path)) > 0.01):
print("ERROR IN POLICY V_COUNT OPTIMIZATION")
log_metabolites = np.append(v_log_counts, f_log_counts)
rxn_flux = max_entropy_functions.oddsDiff(v_log_counts, f_log_counts, mu0,
S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, state)
KQ_f = max_entropy_functions.odds(log_metabolites, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs,
Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r = max_entropy_functions.odds(log_metabolites, mu0, -S_mat, P_mat,
R_back_mat,
delta_increment_for_small_concs,
Keq_inverse, -1)
[RR,
Jac] = max_entropy_functions.calc_Jac2(v_log_counts, f_log_counts, S_mat,
delta_increment_for_small_concs,
KQ_f, KQ_r, state)
A = max_entropy_functions.calc_A(v_log_counts, f_log_counts, S_mat, Jac,
state)
delta_S_metab = max_entropy_functions.calc_deltaS_metab(
v_log_counts, target_v_log_counts)
#delta_S = max_entropy_functions.calc_deltaS(v_log_counts, f_log_counts, S_mat, KQ_f)
[ccc, fcc
] = max_entropy_functions.conc_flux_control_coeff(nvar, A, S_mat,
rxn_flux, RR)
init_action_val = -np.inf
action_value_vec = np.zeros(num_rxns)
state_value_vec = np.zeros(num_rxns)
E_test_vec = np.zeros(num_rxns)
old_E_test_vec = np.zeros(num_rxns)
current_reward_vec = np.zeros(num_rxns)
#print("BEGIN ACTIONS")
for act in range(0, num_rxns):
React_Choice = act #regulate each reaction.
old_E = state[act]
newE = max_entropy_functions.calc_reg_E_step(state, React_Choice, nvar, v_log_counts, f_log_counts,
complete_target_log_counts, S_mat, A, rxn_flux, KQ_f,\
delta_S_metab)
trial_state_sample = state.copy() #DO NOT MODIFY ORIGINAL STATE
trial_state_sample[React_Choice] = newE
states_matrix[:, act] = trial_state_sample.copy()
#re-optimize
new_res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts,
method=Method1,
bounds=(-500, 500),
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat,
P_mat,
delta_increment_for_small_concs,
Keq_constant, trial_state_sample))
if (new_res_lsq.optimality >= 1e-05):
new_res_lsq = least_squares(
max_entropy_functions.derivatives,
v_log_counts,
method=Method2,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant,
trial_state_sample))
if (new_res_lsq.optimality >= 1e-05):
new_res_lsq = least_squares(
max_entropy_functions.derivatives,
v_log_counts,
method=Method3,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant,
trial_state_sample))
new_v_log_counts = new_res_lsq.x
v_log_counts_matrix[:, act] = new_v_log_counts.copy()
new_log_metabolites = np.append(new_v_log_counts, f_log_counts)
KQ_f_matrix[:, act] = max_entropy_functions.odds(
new_log_metabolites, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r_matrix[:, act] = max_entropy_functions.odds(
new_log_metabolites, mu0, -S_mat, P_mat, R_back_mat,
delta_increment_for_small_concs, Keq_inverse, -1)
delta_S_metab_matrix[:, act] = max_entropy_functions.calc_deltaS_metab(
new_v_log_counts, target_v_log_counts)
value_current_state = state_value(
nn_model,
torch.from_numpy(trial_state_sample).float().to(device))
value_current_state = value_current_state.item()
current_reward = reward_value(new_v_log_counts, v_log_counts, \
KQ_f_matrix[:,act], KQ_r_matrix[:,act],\
trial_state_sample, state)
if (current_reward == penalty_exclusion_reward):
rxn_choices.remove(act)
action_value = current_reward + (
gamma) * value_current_state #note, action is using old KQ values
action_value_vec[act] = action_value
old_E_test_vec[act] = old_E
E_test_vec[act] = newE
state_value_vec[act] = value_current_state
current_reward_vec[act] = current_reward #Should have smaller EPR
#print(current_reward)
#USE PENALTY REWARDS
if (len(np.flatnonzero(action_value_vec == action_value_vec.max())) == 0):
print("current action_value_vec")
print(action_value_vec)
print(action_value_vec.max())
#only choose from non penalty rewards
action_choice_index = np.random.choice(
np.flatnonzero(action_value_vec[rxn_choices] ==
action_value_vec[rxn_choices].max()))
action_choice = rxn_choices[action_choice_index]
arr_choice_index = np.flatnonzero(
action_value_vec[rxn_choices] == action_value_vec[rxn_choices].max())
arr_choice = np.asarray(rxn_choices)[arr_choice_index]
arr_choice_reg = np.flatnonzero(state[arr_choice] < 1)
if (arr_choice_reg.size > 1):
print('using tie breaker')
print(arr_choice[arr_choice_reg])
action_choice = np.random.choice(arr_choice[arr_choice_reg])
used_random_step = False
unif_rand = np.random.uniform(0, 1)
if ((unif_rand < epsilon_greedy) and (len(rxn_choices) > 0)):
#if (len(rxn_choices)>1):
# rxn_choices.remove(action_choice)
#print("USING EPSILON GREEDY")
#print(action_choice)
used_random_step = True
random_choice = random.choice(rxn_choices)
action_choice = random_choice
used_random_step = 1
if (current_reward_vec == penalty_exclusion_reward).all():
print("OUT OF REWARDS")
action_choice = -1
if current_reward_vec[action_choice] == penalty_exclusion_reward:
print("state_value_vec")
print(state_value_vec)
print("current_reward_vec")
print(current_reward_vec)
print("used_random_step")
print(used_random_step)
print("rxn_choices")
print(rxn_choices)
# =============================================================================
# print("action_choice")
# print(action_choice)
# print(delta_S_metab_matrix[:,action_choice])
# print(states_matrix[:,action_choice])
# print(states_matrix[:,action_choice]*KQ_f_matrix[:,action_choice])
# print("rewards")
# print(current_reward_vec)
#
# =============================================================================
#breakpoint()
return [action_choice,current_reward_vec[action_choice],\
KQ_f_matrix[:,action_choice],KQ_r_matrix[:,action_choice],\
v_log_counts_matrix[:,action_choice],\
states_matrix[:,action_choice],\
delta_S_metab_matrix[:,action_choice],used_random_step]
def sarsa_n(nn_model, loss_fn, optimizer, scheduler, state_sample, n_back_step,
epsilon_greedy):
#reset for each episode. policy will add
random_steps_taken = 0
nn_steps_taken = 0
maximum_predicted_value = 0
layer_weight = torch.zeros(1, device=device)
final_state = []
final_KQ_f = []
final_KQ_r = []
reached_terminal_state = False
average_loss = []
final_reward = 0
sum_reward_episode = 0
end_of_path = 1000 #this is the maximum length a path can take
states_matrix = np.zeros(shape=(num_rxns, end_of_path + 1))
states_matrix[:, 0] = state_sample
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_static,
method=Method1,
bounds=(-500, 500),
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant, states_matrix[:, 0]))
if (res_lsq.success == False):
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_static,
method=Method2,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs,
Keq_constant, states_matrix[:, 0]))
if (res_lsq.success == False):
res_lsq = least_squares(max_entropy_functions.derivatives,
v_log_counts_static,
method=Method3,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat,
P_mat,
delta_increment_for_small_concs,
Keq_constant, states_matrix[:, 0]))
v_log_counts_current = res_lsq.x.copy()
log_metabolites = np.append(v_log_counts_current, f_log_counts)
rxn_flux_init = max_entropy_functions.oddsDiff(
v_log_counts_current, f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, states_matrix[:, 0])
KQ_f_current = max_entropy_functions.odds(log_metabolites, mu0, S_mat,
R_back_mat, P_mat,
delta_increment_for_small_concs,
Keq_constant)
Keq_inverse = np.power(Keq_constant, -1)
KQ_r_current = max_entropy_functions.odds(log_metabolites, mu0, -S_mat,
P_mat, R_back_mat,
delta_increment_for_small_concs,
Keq_inverse, -1)
delta_S_metab_current = max_entropy_functions.calc_deltaS_metab(
v_log_counts_current, target_v_log_counts)
#[ccc,fcc] = max_entropy_functions.conc_flux_control_coeff(nvar, A_init, S_mat, rxn_flux_init, RR)
reward_vec = np.zeros(end_of_path + 1)
reward_vec[0] = 0.0
rxn_flux_path = rxn_flux_init.copy()
#A_path = A_init.copy()
for t in range(0, end_of_path):
if (t < end_of_path):
#This represents the choice from the current policy.
[React_Choice,reward_vec[t+1],\
KQ_f_current, KQ_r_current,\
v_log_counts_current,\
states_matrix[:,t+1],\
delta_S_metab_current,\
used_random_step] = policy_function(nn_model, states_matrix[:,t], v_log_counts_current, epsilon_greedy)#regulate each reaction.
if (used_random_step):
random_steps_taken += 1
else:
nn_steps_taken += 1
if (React_Choice == -1):
print("out of rewards, final state")
print(states_matrix[:, t + 1])
break
rxn_flux_path = max_entropy_functions.oddsDiff(
v_log_counts_current, f_log_counts, mu0, S_mat, R_back_mat,
P_mat, delta_increment_for_small_concs, Keq_constant,
states_matrix[:, t + 1])
epr_path = max_entropy_functions.entropy_production_rate(
KQ_f_current, KQ_r_current, states_matrix[:, t + 1])
sum_reward_episode += reward_vec[t + 1]
final_state = states_matrix[:, t + 1].copy()
#We stop the path if we have no more positive loss function values, or if we revisit a state.
if ((delta_S_metab_current <= 0.0).all()):
end_of_path = t + 1 #stops simulation at step t+1
reached_terminal_state = True
final_state = states_matrix[:, t + 1].copy()
final_KQ_f = KQ_f_current.copy()
final_KQ_r = KQ_r_current.copy()
final_reward = epr_path
#breakpoint()
print(
"**************************************Path Length ds<0******************************************"
)
print(end_of_path)
print("Final STATE")
print(states_matrix[:, t + 1])
print(rxn_flux_path)
print("original epr")
print(epr_path)
print("all rewards:")
#print(reward_vec[0:t+1])
tau = t - n_back_step + 1
if (tau >= 0):
#THIS IS THE FORWARD
estimate_value = torch.zeros(1, device=device)
for i in range(tau + 1, min(tau + n_back_step, end_of_path) + 1):
estimate_value += (gamma**(i - tau - 1)) * reward_vec[i]
if ((tau + n_back_step) < end_of_path):
value_tau_n = state_value(
nn_model,
torch.from_numpy(
states_matrix[:,
tau + n_back_step]).float().to(device))
estimate_value += (gamma**(n_back_step)) * value_tau_n
value_tau = state_value(
nn_model,
torch.from_numpy(states_matrix[:, tau]).float().to(device))
if (value_tau.requires_grad == False):
print('value tau broken')
if (estimate_value.requires_grad == True):
estimate_value.detach_()
#THIS IS THE END OF FORWARD
#WARNING
#loss ordering should be input with requires_grad == True,
#followed by target with requires_grad == False
optimizer.zero_grad()
loss = (loss_fn(value_tau, estimate_value)) #currently MSE
loss.backward()
clipping_value = 1.0
#torch.nn.utils.clip_grad_value_(nn_model.parameters(), clipping_value)
torch.nn.utils.clip_grad_norm_(nn_model.parameters(),
clipping_value)
optimizer.step()
average_loss.append(loss.item())
if (tau >= (end_of_path - 1)):
break
#after episode is finished, take average loss
average_loss_episode = np.mean(average_loss)
#print(average_loss)
print("index of max error on path")
print(average_loss.index(max(average_loss)))
#print("All rewards")
#print(reward_vec[0:t+1])
return [sum_reward_episode, average_loss_episode,max(average_loss),final_reward, final_state, final_KQ_f,final_KQ_r,\
reached_terminal_state, random_steps_taken,nn_steps_taken]
def policy_function(nn_model, state, v_log_counts_path, *args ):
#last input argument should be epsilon for use when using greedy-epsilon algorithm.
varargin = args
nargin = len(varargin)
epsilon_greedy = 0.0
if (nargin == 1):
epsilon_greedy = varargin[0]
used_random_step=False
rxn_choices = [i for i in range(num_rxns)]
res_lsq = least_squares(max_entropy_functions.derivatives, v_log_counts_path, method=Method1,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, state))
if (res_lsq.success==False):
print("USING DOGBOX")
print("v_log_counts_path")
print(v_log_counts_path)
print("state")
print(state)
res_lsq = least_squares(max_entropy_functions.derivatives, v_log_counts_path, method=Method2,
bounds=(-500,500),xtol=1e-15, args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, state))
if (res_lsq.success==False):
res_lsq = least_squares(max_entropy_functions.derivatives, v_log_counts_path, method=Method3,xtol=1e-15, args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, state))
#v_log_counts = v_log_counts_path.copy()
v_log_counts = res_lsq.x
log_metabolites = np.append(v_log_counts, f_log_counts)
rxn_flux = max_entropy_functions.oddsDiff(v_log_counts, f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, state)
KQ_f = max_entropy_functions.odds(log_metabolites, mu0,S_mat, R_back_mat, P_mat, delta_increment_for_small_concs,Keq_constant);
Keq_inverse = np.power(Keq_constant,-1)
KQ_r = max_entropy_functions.odds(log_metabolites, mu0,-S_mat, P_mat, R_back_mat, delta_increment_for_small_concs,Keq_inverse,-1);
[RR,Jac] = max_entropy_functions.calc_Jac2(v_log_counts, f_log_counts, S_mat, delta_increment_for_small_concs, KQ_f, KQ_r, state)
A = max_entropy_functions.calc_A(v_log_counts, f_log_counts, S_mat, Jac, state )
delta_S_metab = max_entropy_functions.calc_deltaS_metab(v_log_counts, target_v_log_counts)
[ccc,fcc] = max_entropy_functions.conc_flux_control_coeff(nvar, A, S_mat, rxn_flux, RR)
indices = [i for i in range(0,len(Keq_constant))]
action_value_vec = np.zeros(num_rxns)
current_reward_vec = np.zeros(num_rxns)
current_state_vec = np.zeros(num_rxns)
variables=[nn_model,state, nvar, v_log_counts, f_log_counts,\
complete_target_log_counts, A, rxn_flux, KQ_f,\
delta_S_metab,\
mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant]
start = time.time()
with Pool() as pool:
async_result = pool.starmap(potential_step, zip(indices, repeat(variables)))
pool.close()
pool.join()
end = time.time()
total = end-start
#only choose from non penalty rewards
time_cpu=0
time_nn=0
for act in range(0,len(async_result)):
if (async_result[act][1] == penalty_exclusion_reward):
rxn_choices.remove(act)
action_value_vec[act] = async_result[act][0]
current_reward_vec[act] = async_result[act][1]
time_cpu+=async_result[act][7]
time_nn+=async_result[act][8]
current_state_vec[act] = async_result[act][9]
#print(current_reward_vec)
if (len(rxn_choices) == 0):
print("OUT OF REWARDS")
action_choice=-1
else:
try:
action_choice_index = np.random.choice(np.flatnonzero(action_value_vec[rxn_choices] == action_value_vec[rxn_choices].max()))
action_choice = rxn_choices[action_choice_index]
except:
print("WARNING ERROR SHOULD NOT BE HAPPINING")
print("rxn_choices")
print(rxn_choices)
print("action_value_vec")
print(action_value_vec)
print("action_value_vec[rxn_choices].max()")
print(action_value_vec[rxn_choices].max())
print(np.flatnonzero(action_value_vec[rxn_choices] == action_value_vec[rxn_choices].max()))
print("current_reward_vec")
print(current_reward_vec)
print("current_state_vec")
print(current_state_vec)
print("MAXIMUM LAYER WEIGHTS")
for layer in nn_model.modules():
try:
print(torch.max(layer.weight))
except:
print("")
print("async_result")
print(async_result)
action_choice = -1
unif_rand = np.random.uniform(0,1)
if ( (unif_rand < epsilon_greedy) and (len(rxn_choices) > 0)):
#if (len(rxn_choices)>1):
# rxn_choices.remove(action_choice)
#print("USING EPSILON GREEDY")
#print(action_choice)
used_random_step=True
random_choice = random.choice(rxn_choices)
action_choice = random_choice
used_random_step=1
if (np.sum(np.abs(v_log_counts - v_log_counts_path)) > 0.1):
print("ERROR IN POLICY V_COUNT OPTIMIZATION")
#print("async_result")
#print(async_result)
print("state")
print(state)
print("v_log_counts")
print(v_log_counts)
print("v_log_counts_path")
print(v_log_counts_path)
print("current_reward_vec")
print(current_reward_vec)
print("action_value_vec")
print(action_value_vec)
print("rxn_choices")
print(rxn_choices)
print("MAXIMUM LAYER WEIGHTS")
for layer in nn_model.modules():
try:
print(torch.max(layer.weight))
except:
print("")
#async_result order
#[action_value, current_reward,KQ_f_new,KQ_r_new,new_v_log_counts,trial_state_sample,new_delta_S_metab]
return [action_choice,async_result[action_choice][1],\
async_result[action_choice][2],async_result[action_choice][3],\
async_result[action_choice][4],\
async_result[action_choice][5],\
async_result[action_choice][6],used_random_step,time_cpu,time_nn]
def potential_step(index, other_args):
React_Choice=index
nn_model,state, nvar, v_log_counts, f_log_counts,\
complete_target_log_counts, A, rxn_flux, KQ_f,\
delta_S_metab,\
mu0, S_mat, R_back_mat, P_mat, \
delta_increment_for_small_concs, Keq_constant = other_args
newE = max_entropy_functions.calc_reg_E_step(state, React_Choice, nvar, v_log_counts, f_log_counts,
complete_target_log_counts, S_mat, A, rxn_flux, KQ_f,\
delta_S_metab)
trial_state_sample = state.copy()#DO NOT MODIFY ORIGINAL STATE
trial_state_sample[React_Choice] = newE
#re-optimize
start_cpu = time.time()
new_res_lsq = least_squares(max_entropy_functions.derivatives, v_log_counts, method=Method1,
xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, trial_state_sample))
if (new_res_lsq.success==False):
print("USING DOGBOX")
print("v_log_counts")
print(v_log_counts)
print("trial_state_sample")
print(trial_state_sample)
new_res_lsq = least_squares(max_entropy_functions.derivatives, v_log_counts, method=Method2,
bounds=(-500,500), xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, trial_state_sample))
if (new_res_lsq.success==False):
new_res_lsq = least_squares(max_entropy_functions.derivatives, v_log_counts, method=Method3,xtol=1e-15,
args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat,
delta_increment_for_small_concs, Keq_constant, trial_state_sample))
end_cpu = time.time()
new_v_log_counts = new_res_lsq.x
new_log_metabolites = np.append(new_v_log_counts, f_log_counts)
new_delta_S_metab = max_entropy_functions.calc_deltaS_metab(new_v_log_counts, target_v_log_counts)
KQ_f_new = max_entropy_functions.odds(new_log_metabolites, mu0,S_mat, R_back_mat, P_mat, delta_increment_for_small_concs,Keq_constant);
Keq_inverse = np.power(Keq_constant,-1)
KQ_r_new = max_entropy_functions.odds(new_log_metabolites, mu0,-S_mat, P_mat, R_back_mat, delta_increment_for_small_concs,Keq_inverse,-1);
begin_nn = time.time()
value_current_state = state_value(nn_model, torch.from_numpy(trial_state_sample).float().to(device) )
value_current_state = value_current_state.item()
end_nn = time.time()
current_reward = reward_value(new_v_log_counts, v_log_counts, \
KQ_f_new, KQ_r_new,\
trial_state_sample, state)
action_value = current_reward + (gamma) * value_current_state #note, action is using old KQ values
return [action_value, current_reward,KQ_f_new,KQ_r_new,new_v_log_counts,trial_state_sample,new_delta_S_metab, end_cpu-start_cpu,end_nn-begin_nn,value_current_state]
def run(argv):
try:
os.makedirs(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/data')
except FileExistsError:
# directory already exists
pass
try:
os.makedirs(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data')
except FileExistsError:
# directory already exists
pass
#default values
#If no experimental data is available, we can estimate using 'rule-of-thumb' data at 0.001
use_experimental_data=False
learning_rate=1e-8 #3rd
epsilon=0.05 #4th
eps_threshold=25 #5th
gamma = 0.9 #6th
updates = 500
penalty_reward_scalar=0.0
#load input
total = len(sys.argv)
cmdargs = str(sys.argv)
print ("The total numbers of args passed to the script: %d " % total)
print ("Args list: %s " % cmdargs)
print ("Script name: %s" % str(sys.argv[0]))
for i in range(total):
print ("Argument # %d : %s" % (i, str(sys.argv[i])))
sim_number=int(sys.argv[1])
n_back_step=int(sys.argv[2])
if (total > 3):
use_experimental_data=bool(int(sys.argv[3]))
if (total > 4):
learning_rate=float(sys.argv[4])
if (total > 5):
epsilon=float(sys.argv[5])
if (total > 6):
eps_threshold=float(sys.argv[6])
if (total > 7):
gamma=float(sys.argv[7])
pd.set_option('display.max_columns', None,'display.max_rows', None)
print("sim")
print(sim_number)
print("n_back_step")
print(n_back_step)
print("using experimental metabolite data")
print(use_experimental_data)
print("learning_rate")
print(learning_rate)
print("epsilon")
print(epsilon)
print("eps_threshold")
print(eps_threshold)
print("gamma")
print(gamma)
T = 298.15
R = 8.314e-03
RT = R*T
N_avogadro = 6.022140857e+23
VolCell = 1.0e-15
Concentration2Count = N_avogadro * VolCell
concentration_increment = 1/(N_avogadro*VolCell)
np.set_printoptions(suppress=True)#turn off printin
# In[3]:
#with open( cwd + '/TCA_PPP_GLYCOLYSIS_CELLWALL/TCA_PPP_Glycolysis_CellWall3b.dat', 'r') as f:
# print(f.read())
# In[5]:
fdat = open(cwd + '/TCA_PPP_GLYCOLYSIS_CELLWALL/TCA_PPP_Glycolysis_CellWall3b.dat', 'r')
#fdat = open('TCA_PPP_Glycolysis.dat', 'r')
left ='LEFT'
right = 'RIGHT'
left_compartment = 'LEFT_COMPARTMENT'
right_compartment = 'RIGHT_COMPARTMENT'
enzyme_level = 'ENZYME_LEVEL'
deltag0 = 'DGZERO'
deltag0_sigma = 'DGZERO StdDev'
same_compartment = 'Same Compartment?'
full_rxn = 'Full Rxn'
reactions = pd.DataFrame(index=[],columns=[left, right, left_compartment, right_compartment, enzyme_level, deltag0, deltag0_sigma, same_compartment,full_rxn])
reactions.index.name='REACTION'
S_matrix = pd.DataFrame(index=[],columns=[enzyme_level])
S_matrix.index.name='REACTION'
for line in fdat:
if (line.startswith('REACTION')):
rxn_name = line[9:-1].lstrip()
S_matrix.loc[rxn_name,enzyme_level] = 1.0
reactions.loc[rxn_name,enzyme_level] = 1.0
if (re.match("^LEFT\s",line)):
line = line.upper()
left_rxn = line[4:-1].lstrip()
left_rxn = re.sub(r'\s+$', '', left_rxn) #Remove trailing white space
reactions.loc[rxn_name,left] = left_rxn
elif (re.match('^RIGHT\s',line)):
line = line.upper()
right_rxn = line[5:-1].lstrip()
right_rxn = re.sub(r'\s+$', '', right_rxn) #Remove trailing white space
reactions.loc[rxn_name,right] = right_rxn
elif (line.startswith(left_compartment)):
cpt_name = line[16:-1].lstrip()
reactions.loc[rxn_name,left_compartment] = cpt_name
reactants = re.split(' \+ ',left_rxn)
for idx in reactants:
values = re.split(' ', idx);
if len(values) == 2:
stoichiometry = np.float64(values[0]);
molecule = values[1];
if not re.search(':',molecule):
molecule = molecule + ':' + cpt_name
else:
stoichiometry = np.float64(-1.0);
molecule = values[0];
if not re.search(':',molecule):
molecule = molecule + ':' + cpt_name
S_matrix.loc[rxn_name,molecule] = stoichiometry;
elif (line.startswith(right_compartment)):
cpt_name = line[17:-1].lstrip()
reactions.loc[rxn_name,right_compartment] = cpt_name
products = re.split(' \+ ',right_rxn)
for idx in products:
values = re.split(' ', idx);
if len(values) == 2:
stoichiometry = np.float64(values[0]);
molecule = values[1];
if not re.search(':',molecule):
molecule = molecule + ':' + cpt_name
else:
stoichiometry = np.float64(1.0);
molecule = values[0];
if not re.search(':',molecule):
molecule = molecule + ':' + cpt_name
S_matrix.loc[rxn_name,molecule] = stoichiometry;
elif (re.match("^ENZYME_LEVEL\s", line)):
level = line[12:-1].lstrip()
reactions.loc[rxn_name,enzyme_level] = float(level)
S_matrix.loc[rxn_name,enzyme_level] = float(level)
elif re.match('^COMMENT',line):
continue
elif re.match(r'//',line):
continue
elif re.match('^#',line):
continue
# elif (re.match("^[N,P]REGULATION\s", line)):
# reg = line
# reactions.loc[rxn_name,regulation] = reg
fdat.close()
S_matrix.fillna(0,inplace=True)
S_active = S_matrix[S_matrix[enzyme_level] > 0.0]
active_reactions = reactions[reactions[enzyme_level] > 0.0]
del S_active[enzyme_level]
# Delete any columns/metabolites that have all zeros in the S matrix:
S_active = S_active.loc[:, (S_active != 0).any(axis=0)]
np.shape(S_active.values)
#print(S_active.shape)
#print(S_active)
reactions[full_rxn] = reactions[left] + ' = ' + reactions[right]
# In[6]:
if (1):
for idx in reactions.index:
#print(idx,flush=True)
boltzmann_rxn_str = reactions.loc[idx,'Full Rxn']
if re.search(':',boltzmann_rxn_str):
all_cmprts = re.findall(':\S+', boltzmann_rxn_str)
[s.replace(':', '') for s in all_cmprts] # remove all the ':'s
different_compartments = 0
for cmpt in all_cmprts:
if not re.match(all_cmprts[0],cmpt):
different_compartments = 1
if ((not different_compartments) and (reactions[left_compartment].isnull or reactions[right_compartment].isnull)):
reactions.loc[idx,left_compartment] = cmpt
reactions.loc[idx,right_compartment] = cmpt
reactions.loc[idx,same_compartment] = True
if different_compartments:
reactions.loc[idx,same_compartment] = False
else:
if (reactions.loc[idx,left_compartment] == reactions.loc[idx,right_compartment]):
reactions.loc[idx,same_compartment] = True
else:
reactions.loc[idx,same_compartment] = False
#print(reactions)
reactions.loc['CSm',deltag0] = -35.8057
reactions.loc['ACONTm',deltag0] = 7.62962
reactions.loc['ICDHxm',deltag0] = -2.6492
reactions.loc['AKGDam',deltag0] = -37.245
reactions.loc['SUCOASm',deltag0] = 2.01842
reactions.loc['SUCD1m',deltag0] = -379.579
reactions.loc['FUMm',deltag0] = -3.44728
reactions.loc['MDHm',deltag0] = 29.5419
reactions.loc['GAPD',deltag0] = 5.24202
reactions.loc['PGK',deltag0] = -18.5083
reactions.loc['TPI',deltag0] = 5.49798
reactions.loc['FBA',deltag0] = 21.4506
reactions.loc['PYK',deltag0] = -27.3548
reactions.loc['PGM',deltag0] = 4.17874
reactions.loc['ENO',deltag0] = -4.0817
reactions.loc['HEX1',deltag0] = -16.7776
reactions.loc['PGI',deltag0] = 2.52206
reactions.loc['PFK',deltag0] = -16.1049
reactions.loc['PYRt2m',deltag0] = -RT*np.log(10)
reactions.loc['PDHm',deltag0] = -44.1315
reactions.loc['G6PDH2r',deltag0] = -3.89329
reactions.loc['PGL',deltag0] = -22.0813
reactions.loc['GND',deltag0] = 2.32254
reactions.loc['RPE',deltag0] = -3.37
reactions.loc['RPI',deltag0] = -1.96367
reactions.loc['TKT2',deltag0] = -10.0342
reactions.loc['TALA',deltag0] = -0.729232
reactions.loc['FBA3',deltag0] = 13.9499
reactions.loc['PFK_3',deltag0] = -9.33337
reactions.loc['TKT1',deltag0] = -3.79303
reactions.loc['Glutamine-fructose-6-phosphate aminotransferase',deltag0] = -13.4054
reactions.loc['Glucosamine-6-phosphate N-acetyltransferase',deltag0] = -23.7065
reactions.loc['N-acetylglucosamine-phosphate mutase',deltag0] = 4.65558
reactions.loc['UDP N-acetylglucosamine pyrophosphorylase',deltag0] = 0.539147
reactions.loc['Hyaluronan Synthase',deltag0] = -14.4143
reactions.loc['Phosphoglucomutase',deltag0] = 7.41831
reactions.loc['UTP-glucose-1-phosphate uridylyltransferase',deltag0] = 1.51043
reactions.loc['1,3-beta-glucan synthase',deltag0] = -11.534
reactions.loc['Citrate-oxaloacetate exchange',deltag0] = 0
reactions.loc['CITRATE_LYASE',deltag0] = 10.0299
reactions.loc['MDHc',deltag0] = -29.5419
reactions.loc['MDH-NADPc',deltag0] = -29.7376
reactions.loc['ME1c',deltag0] = 4.56191
reactions.loc['ME2c',deltag0] = 4.75763
reactions.loc['Pyruvate Carboxylase',deltag0] = -0.795825
reactions.loc['Aldose 1-epimerase',deltag0] = 0
reactions.loc['HEX1a',deltag0] = -16.7776
reactions.loc['PGI-1',deltag0] = 2.52206
reactions.loc['CSm',deltag0_sigma] = 0.930552
reactions.loc['ACONTm',deltag0_sigma] = 0.733847
reactions.loc['ICDHxm',deltag0_sigma] = 7.62095
reactions.loc['AKGDam',deltag0_sigma] = 7.97121
reactions.loc['SUCOASm',deltag0_sigma] = 1.48197
reactions.loc['SUCD1m',deltag0_sigma] = 7.8098
reactions.loc['FUMm',deltag0_sigma] = 0.607693
reactions.loc['MDHm',deltag0_sigma] = 0.422376
reactions.loc['GAPD',deltag0_sigma] = 0.895659
reactions.loc['PGK',deltag0_sigma] = 0.889982
reactions.loc['TPI',deltag0_sigma] = 0.753116
reactions.loc['FBA',deltag0_sigma] = 0.87227
reactions.loc['PYK',deltag0_sigma] = 0.939774
reactions.loc['PGM',deltag0_sigma] = 0.65542
reactions.loc['ENO',deltag0_sigma] = 0.734193
reactions.loc['HEX1',deltag0_sigma] = 0.715237
reactions.loc['PGI',deltag0_sigma] = 0.596775
reactions.loc['PFK',deltag0_sigma] = 0.886629
reactions.loc['PYRt2m',deltag0_sigma] = 0
reactions.loc['PDHm',deltag0_sigma] = 7.66459
reactions.loc['G6PDH2r',deltag0_sigma] = 2.11855
reactions.loc['PGL',deltag0_sigma] = 2.62825
reactions.loc['GND',deltag0_sigma] = 7.60864
reactions.loc['RPE',deltag0_sigma] = 1.16485
reactions.loc['RPI',deltag0_sigma] = 1.16321
reactions.loc['TKT2',deltag0_sigma] = 2.08682
reactions.loc['TALA',deltag0_sigma] = 1.62106
reactions.loc['FBA3',deltag0_sigma] = 7.36854
reactions.loc['PFK_3',deltag0_sigma] = 7.3671
reactions.loc['TKT1',deltag0_sigma] = 2.16133
reactions.loc['Glutamine-fructose-6-phosphate aminotransferase',deltag0_sigma] = 3.08807
reactions.loc['Glucosamine-6-phosphate N-acetyltransferase',deltag0_sigma] = 4.26738
reactions.loc['N-acetylglucosamine-phosphate mutase',deltag0_sigma] = 3.06369
reactions.loc['UDP N-acetylglucosamine pyrophosphorylase',deltag0_sigma] = 3.12527
reactions.loc['Hyaluronan Synthase',deltag0_sigma] = 9.46851
reactions.loc['Phosphoglucomutase',deltag0_sigma] = 1.09029
reactions.loc['UTP-glucose-1-phosphate uridylyltransferase',deltag0_sigma] = 1.14644
reactions.loc['1,3-beta-glucan synthase',deltag0_sigma] = 7.80447
reactions.loc['Citrate-oxaloacetate exchange',deltag0_sigma] = 0
reactions.loc['CITRATE_LYASE',deltag0_sigma] = 0.928303
reactions.loc['MDHc',deltag0_sigma] = 0.422376
reactions.loc['MDH-NADPc',deltag0_sigma] = 0.531184
reactions.loc['ME1c',deltag0_sigma] = 7.60174
reactions.loc['ME2c',deltag0_sigma] = 7.61042
reactions.loc['Pyruvate Carboxylase',deltag0_sigma] = 7.60419
reactions.loc['Aldose 1-epimerase',deltag0_sigma] = 0
reactions.loc['HEX1a',deltag0_sigma] = 0.715237
reactions.loc['PGI-1',deltag0_sigma] = 0.596775
# ## Calculate Standard Free Energies of Reaction
# In[49]:
conc = 'Conc'
variable = 'Variable'
conc_exp = 'Conc_Experimental'
metabolites = pd.DataFrame(index = S_active.columns, columns=[conc,conc_exp,variable])
metabolites[conc] = 0.001
metabolites[variable] = True
# Set the fixed metabolites:
metabolites.loc['ATP:MITOCHONDRIA',conc] = 9.600000e-03
metabolites.loc['ATP:MITOCHONDRIA',variable] = False
metabolites.loc['ADP:MITOCHONDRIA',conc] = 5.600000e-04
metabolites.loc['ADP:MITOCHONDRIA',variable] = False
metabolites.loc['ORTHOPHOSPHATE:MITOCHONDRIA',conc] = 2.000000e-02
metabolites.loc['ORTHOPHOSPHATE:MITOCHONDRIA',variable] = False
metabolites.loc['ATP:CYTOSOL',conc] = 9.600000e-03
metabolites.loc['ATP:CYTOSOL',variable] = False
metabolites.loc['ADP:CYTOSOL',conc] = 5.600000e-04
metabolites.loc['ADP:CYTOSOL',variable] = False
metabolites.loc['ORTHOPHOSPHATE:CYTOSOL',conc] = 2.000000e-02
metabolites.loc['ORTHOPHOSPHATE:CYTOSOL',variable] = False
metabolites.loc['UTP:CYTOSOL',conc] = 9.600000e-03
metabolites.loc['UTP:CYTOSOL',variable] = False
metabolites.loc['UDP:CYTOSOL',conc] = 5.600000e-04
metabolites.loc['UDP:CYTOSOL',variable] = False
metabolites.loc['DIPHOSPHATE:CYTOSOL',conc] = 2.000000e-02
metabolites.loc['DIPHOSPHATE:CYTOSOL',variable] = False
metabolites.loc['NADH:MITOCHONDRIA',conc] = 8.300000e-05
metabolites.loc['NADH:MITOCHONDRIA',variable] = False
metabolites.loc['NAD+:MITOCHONDRIA',conc] = 2.600000e-03
metabolites.loc['NAD+:MITOCHONDRIA',variable] = False
metabolites.loc['NADH:CYTOSOL',conc] = 8.300000e-05
metabolites.loc['NADH:CYTOSOL',variable] = False
metabolites.loc['NAD+:CYTOSOL',conc] = 2.600000e-03
metabolites.loc['NAD+:CYTOSOL',variable] = False
metabolites.loc['NADPH:CYTOSOL',conc] = 8.300000e-05 #also use 1.2e-4
metabolites.loc['NADPH:CYTOSOL',variable] = False
metabolites.loc['NADP+:CYTOSOL',conc] = 2.600000e-03 #also use 2.1e-6
metabolites.loc['NADP+:CYTOSOL',variable] = False
metabolites.loc['COA:MITOCHONDRIA',conc] = 1.400000e-03
metabolites.loc['COA:MITOCHONDRIA',variable] = False
metabolites.loc['COA:CYTOSOL',conc] = 1.400000e-03
metabolites.loc['COA:CYTOSOL',variable] = False
metabolites.loc['CO2:MITOCHONDRIA',conc] = 1.000000e-04
metabolites.loc['CO2:MITOCHONDRIA',variable] = False
metabolites.loc['CO2:CYTOSOL',conc] = 1.000000e-04
metabolites.loc['CO2:CYTOSOL',variable] = False
metabolites.loc['H2O:MITOCHONDRIA',conc] = 55.5
metabolites.loc['H2O:MITOCHONDRIA',variable] = False
metabolites.loc['H2O:CYTOSOL',conc] = 55.5
metabolites.loc['H2O:CYTOSOL',variable] = False
metabolites.loc['BETA-D-GLUCOSE:CYTOSOL',conc] = 2.0e-03
metabolites.loc['BETA-D-GLUCOSE:CYTOSOL',variable] = False
metabolites.loc["CHITOBIOSE:CYTOSOL",conc] = 2.0e-09
metabolites.loc["CHITOBIOSE:CYTOSOL",variable] = False
metabolites.loc['1,3-BETA-D-GLUCAN:CYTOSOL',conc] = 2.0e-09
metabolites.loc['1,3-BETA-D-GLUCAN:CYTOSOL',variable] = False
metabolites.loc['L-GLUTAMINE:CYTOSOL',conc] = 2.0e-03
metabolites.loc['L-GLUTAMINE:CYTOSOL',variable] = False
metabolites.loc['L-GLUTAMATE:CYTOSOL',conc] = 2.0e-04
metabolites.loc['L-GLUTAMATE:CYTOSOL',variable] = False
metabolites.loc['CELLOBIOSE:CYTOSOL',conc] = 2.0e-04
metabolites.loc['CELLOBIOSE:CYTOSOL',variable] = False
metabolites.loc['N-ACETYL-D-GLUCOSAMINE:CYTOSOL',conc] = 1.0e-08
metabolites.loc['N-ACETYL-D-GLUCOSAMINE:CYTOSOL',variable] = False
#When loading experimental concentrations, first copy current
#rule of thumb then overwrite with data values.
metabolites[conc_exp] = metabolites[conc]
metabolites.loc['2-OXOGLUTARATE:MITOCHONDRIA',conc_exp] = 0.0000329167257825644
metabolites.loc['ISOCITRATE:MITOCHONDRIA',conc_exp] = 0.000102471198594958
metabolites.loc['PHOSPHOENOLPYRUVATE:CYTOSOL',conc_exp] = 0.0000313819870767023
metabolites.loc['D-GLYCERALDEHYDE-3-PHOSPHATE:CYTOSOL',conc_exp] = 0.0000321630949358949
metabolites.loc['FUMARATE:MITOCHONDRIA',conc_exp] = 0.00128926137523035
metabolites.loc['L-GLUTAMINE:CYTOSOL',conc_exp] = 0.0034421144256392
metabolites.loc['PYRUVATE:MITOCHONDRIA',conc_exp] = 0.0000778160985710288
metabolites.loc['PYRUVATE:CYTOSOL',conc_exp] = 0.0000778160985710288
metabolites.loc['D-FRUCTOSE_6-PHOSPHATE:CYTOSOL',conc_exp] = 0.00495190614473117
metabolites.loc['D-RIBOSE-5-PHOSPHATE:CYTOSOL',conc_exp] = 0.0000849533575412862
metabolites.loc['CITRATE:MITOCHONDRIA',conc_exp] = 0.000485645834537379
metabolites.loc['CITRATE:CYTOSOL',conc_exp] = 0.000485645834537379
metabolites.loc['(S)-MALATE:MITOCHONDRIA',conc_exp] = 0.00213827060541153
metabolites.loc['(S)-MALATE:CYTOSOL',conc_exp] = 0.00213827060541153
metabolites.loc['SEDOHEPTULOSE_7-PHOSPHATE:CYTOSOL',conc_exp] = 0.00203246193132095
metabolites.loc['D-RIBULOSE-5-PHOSPHATE:CYTOSOL',conc_exp] = 0.000468439334729429
metabolites.loc['L-GLUTAMATE:CYTOSOL',conc_exp] = 0.00557167476932484
metabolites.loc['SUCCINATE:MITOCHONDRIA',conc_exp] = 0.000942614767220802
metabolites.loc['D-XYLULOSE-5-PHOSPHATE:CYTOSOL',conc_exp] = 0.000468439334729429
nvariables = metabolites[metabolites[variable]].count()
nvar = nvariables[variable]
metabolites.sort_values(by=variable, axis=0,ascending=False, inplace=True,)
#print(metabolites)
#%%
nvariables = metabolites[metabolites[variable]].count()
nvar = nvariables[variable]
metabolites.sort_values(by=variable, axis=0,ascending=False, inplace=True,)
#print(metabolites)
# ## Prepare model for optimization
# - Adjust S Matrix to use only reactions with activity > 0, if necessary.
# - Water stoichiometry in the stiochiometric matrix needs to be set to zero since water is held constant.
# - The initial concentrations of the variable metabolites are random.
# - All concentrations are changed to log counts.
# - Equilibrium constants are calculated from standard free energies of reaction.
# - R (reactant) and P (product) matrices are derived from S.
# Make sure all the indices and columns are in the correct order:
active_reactions = reactions[reactions[enzyme_level] > 0.0]
#print(reactions)
#print(metabolites.index)
Sactive_index = S_active.index
active_reactions.reindex(index = Sactive_index, copy = False)
S_active = S_active.reindex(columns = metabolites.index, copy = False)
S_active['H2O:MITOCHONDRIA'] = 0
S_active['H2O:CYTOSOL'] = 0
#####################################
#####################################
#THIS IS MAKING FLUX -> 0.0
where_are_NaNs = np.isnan(S_active)
S_active[where_are_NaNs] = 0
#print(S_active[:])
S_mat = S_active.values
Keq_constant = np.exp(-active_reactions[deltag0].astype('float')/RT)
#print(Keq_constant)
Keq_constant = Keq_constant.values
P_mat = np.where(S_mat>0,S_mat,0)
R_back_mat = np.where(S_mat<0, S_mat, 0)
E_regulation = np.ones(Keq_constant.size) # THis is the vector of enzyme activities, Range: 0 to 1.
mu0 = 1 #Dummy parameter for now; reserved for free energies of formation
#If no experimental data is available, we can estimate using 'rule-of-thumb' data at 0.001
conc_type=conc
if (use_experimental_data):
print("USING EXPERIMENTAL DATA")
conc_type=conc_exp
variable_concs = np.array(metabolites[conc_type].iloc[0:nvar].values, dtype=np.float64)
v_log_concs = -10 + 10*np.random.rand(nvar) #Vary between 1 M to 1.0e-10 M
v_concs = np.exp(v_log_concs)
v_log_counts_stationary = np.log(v_concs*Concentration2Count)
v_log_counts = v_log_counts_stationary
#print(v_log_counts)
fixed_concs = np.array(metabolites[conc_type].iloc[nvar:].values, dtype=np.float64)
fixed_counts = fixed_concs*Concentration2Count
f_log_counts = np.log(fixed_counts)
complete_target_log_counts = np.log(Concentration2Count * metabolites[conc_type].values)
target_v_log_counts = complete_target_log_counts[0:nvar]
target_f_log_counts = complete_target_log_counts[nvar:]
#WARNING:::::::::::::::CHANGE BACK TO ZEROS
delta_increment_for_small_concs = (10**-50)*np.zeros(metabolites[conc_type].values.size);
variable_concs_begin = np.array(metabolites[conc_type].iloc[0:nvar].values, dtype=np.float64)
#%% Basic test
v_log_counts = np.log(variable_concs_begin*Concentration2Count)
#r_log_counts = -10 + 10*np.random.rand(v_log_counts.size)
#v_log_counts = r_log_counts
#print('====== Without adjusting Keq_constant ======')
E_regulation = np.ones(Keq_constant.size) # THis is the vector of enzyme activities, Range: 0 to 1.
nvar = v_log_counts.size
#WARNING: INPUT LOG_COUNTS TO ALL FUNCTIONS. CONVERSION TO COUNTS IS DONE INTERNALLY
res_lsq1 = least_squares(max_entropy_functions.derivatives, v_log_counts, method='lm',xtol=1e-15, args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, E_regulation))
if (res_lsq1.success==False):
res_lsq1 = least_squares(max_entropy_functions.derivatives, v_log_counts,method='dogbox',xtol=1e-15, args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, E_regulation))
if (res_lsq1.success==False):
res_lsq1 = least_squares(max_entropy_functions.derivatives, v_log_counts,method='trf',xtol=1e-15, args=(f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, E_regulation))
rxn_flux = max_entropy_functions.oddsDiff(res_lsq1.x, f_log_counts, mu0, S_mat, R_back_mat, P_mat, delta_increment_for_small_concs, Keq_constant, E_regulation)
# In[ ]:
begin_log_metabolites = np.append(res_lsq1.x,f_log_counts)
##########################################
##########################################
#####################TESTER###############
E_regulation = np.ones(Keq_constant.size) # THis is the vector of enzyme activities, Range: 0 to 1.
log_metabolites = np.append(res_lsq1.x,f_log_counts)
KQ_f = max_entropy_functions.odds(log_metabolites,mu0,S_mat, R_back_mat, P_mat, delta_increment_for_small_concs,Keq_constant)
Keq_inverse = np.power(Keq_constant,-1)
KQ_r = max_entropy_functions.odds(log_metabolites,mu0,-S_mat, P_mat, R_back_mat, delta_increment_for_small_concs,Keq_inverse,-1)
[RR,Jac] = max_entropy_functions.calc_Jac2(res_lsq1.x, f_log_counts, S_mat, delta_increment_for_small_concs, KQ_f, KQ_r, E_regulation)
A = max_entropy_functions.calc_A(res_lsq1.x,f_log_counts, S_mat, Jac, E_regulation )
[ccc,fcc] = max_entropy_functions.conc_flux_control_coeff(nvar, A, S_mat, rxn_flux, RR)
React_Choice=6
newE = max_entropy_functions.calc_reg_E_step(E_regulation,React_Choice, nvar, res_lsq1.x, f_log_counts, complete_target_log_counts,
S_mat, A, rxn_flux,KQ_f)
delta_S_metab = max_entropy_functions.calc_deltaS_metab(res_lsq1.x, target_v_log_counts);
ipolicy = 7 #use ipolicy=1 or 4
reaction_choice = max_entropy_functions.get_enzyme2regulate(ipolicy, delta_S_metab, ccc, KQ_f, E_regulation, res_lsq1.x)
#%%
#device = torch.device("cpu")
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print('Using device:', device)
#set variables in ML program
me.device=device
me.v_log_counts_static = v_log_counts_stationary
me.target_v_log_counts = target_v_log_counts
me.complete_target_log_counts = complete_target_log_counts
me.Keq_constant = Keq_constant
me.f_log_counts = f_log_counts
me.P_mat = P_mat
me.R_back_mat = R_back_mat
me.S_mat = S_mat
me.delta_increment_for_small_concs = delta_increment_for_small_concs
me.nvar = nvar
me.mu0 = mu0
me.gamma = gamma
me.num_rxns = Keq_constant.size
me.penalty_reward_scalar=penalty_reward_scalar
#%%
N, D_in, H, D_out = 1, Keq_constant.size, 20*Keq_constant.size, 1
# Create random Tensors to hold inputs and outputs
x = torch.rand(1000, D_in, device=device)
nn_model = torch.nn.Sequential(
torch.nn.Linear(D_in, H),
torch.nn.Tanh(),
torch.nn.Linear(H,D_out)).to(device)
loss_fn = torch.nn.MSELoss(reduction='sum')
#learning_rate=5e-6
#optimizer = torch.optim.SGD(nn_model.parameters(), lr=learning_rate, momentum=0.9)
optimizer = torch.optim.SGD(nn_model.parameters(), lr=learning_rate, momentum=0.9)
#optimizer = torch.optim.Adam(nn_model.parameters(), lr=3e-4)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, patience=200, verbose=True, min_lr=1e-10,cooldown=10,threshold=1e-5)
#scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, patience=100, verbose=True, min_lr=1e-10,cooldown=10,threshold=1e-4)
#%% SGD UPDATE TEST
#attempted iterations to update theta_linear
v_log_counts = v_log_counts_stationary.copy()
episodic_loss = []
episodic_loss_max = []
episodic_epr = []
episodic_reward = []
episodic_nn_step = []
episodic_random_step = []
epsilon_greedy_init = epsilon
final_states=np.zeros(Keq_constant.size)
final_KQ_fs=np.zeros(Keq_constant.size)
final_KQ_rs=np.zeros(Keq_constant.size)
epr_per_state=[]
was_state_terminal=[]
for update in range(0,updates):
x_changing = 1*torch.rand(1000, D_in, device=device)
#generate state to use
state_sample = np.zeros(Keq_constant.size)
for sample in range(0,len(state_sample)):
state_sample[sample] = np.random.uniform(1,1)
#annealing test
if ((update % eps_threshold== 0) and (update != 0)):
epsilon=epsilon/2
print("RESET epsilon ANNEALING")
print(epsilon)
prediction_x_changing_previous = nn_model(x_changing)
#nn_model.train()
[sum_reward, average_loss,max_loss,final_epr,final_state,final_KQ_f,final_KQ_r, reached_terminal_state,\
random_steps_taken,nn_steps_taken] = me.sarsa_n(nn_model,loss_fn, optimizer, scheduler, state_sample, n_back_step, epsilon)
print("EPISODE")
print(update)
print("MAXIMUM LAYER WEIGHTS")
for layer in nn_model.modules():
try:
print(torch.max(layer.weight))
except:
print("")
print('random,nn steps')
print(random_steps_taken)
print(nn_steps_taken)
if (reached_terminal_state):
was_state_terminal.append(1)
else:
was_state_terminal.append(0)
final_states = np.vstack((final_states,final_state))
final_KQ_fs = np.vstack((final_KQ_fs,final_KQ_f))
final_KQ_rs = np.vstack((final_KQ_rs,final_KQ_r))
epr_per_state.append(final_epr)
episodic_epr.append(final_epr)
episodic_loss.append(average_loss)
episodic_loss_max.append(max_loss)
episodic_reward.append(sum_reward)
episodic_nn_step.append(nn_steps_taken)
episodic_random_step.append(random_steps_taken)
scheduler.step(average_loss)
print("TOTAL REWARD")
print(sum_reward)
print("ave loss")
print(average_loss)
print("max_loss")
print(max_loss)
print(optimizer.state_dict)
print(scheduler.state_dict())
prediction_x_changing = nn_model(x_changing)
total_prediction_changing_diff = sum(abs(prediction_x_changing - prediction_x_changing_previous))
print("TOTALPREDICTION")
print(total_prediction_changing_diff)
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/data/'+
'temp_episodic_loss_'+str(n_back_step) +
'_lr'+str(learning_rate)+
'_'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_'+str(sim_number)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'.txt', episodic_loss, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/data/'+
'temp_epr_'+str(n_back_step) +
'_lr'+str(learning_rate)+
'_'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_'+str(sim_number)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'.txt', episodic_epr, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/data/'+
'temp_episodic_reward_'+str(n_back_step)+
'_lr'+str(learning_rate)+
'_'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+'_'+str(sim_number)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'.txt', episodic_reward, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/data/'+
'temp_final_states_'+str(n_back_step)+
'_lr'+str(learning_rate)+
'_'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+'_'+str(sim_number)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'.txt', final_states, fmt='%f')
if (update > 200):
if ((max(episodic_loss[-100:])-min(episodic_loss[-100:]) < 0.025) and (update > 350)):
break
#%%
#gamma9 -> gamma=0.9
#n8 -> n_back_step=8
#k5 -> E=E-E/5 was used
#lr5e6 -> begin lr=0.5*e-6
torch.save(nn_model, cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'complete_model_gly_tca_gog_gamma9_n'+str(n_back_step)+'_k5_'\
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_' +str(int(use_experimental_data))+
'_sim'+str(sim_number) + '.pth')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'episodic_terminal_state_gamma9_n'+str(n_back_step)+'_k5_'
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+
'.txt', was_state_terminal, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'episodic_loss_gamma9_n'+str(n_back_step)+'_k5_'
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+
'.txt', episodic_loss, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'episodic_loss_max_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+
'.txt', episodic_loss_max, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'episodic_reward_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+
'.txt', episodic_reward, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'final_states_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+\
'.txt', final_states, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'final_KQF_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+\
'.txt', final_KQ_fs, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'final_KQR_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_'+str(int(use_experimental_data))+
'_sim'+str(sim_number)+\
'.txt', final_KQ_rs, fmt='%f')
np.savetxt(cwd+'/TCA_PPP_GLYCOLYSIS_CELLWALL/models_final_data/'+
'epr_per_state_gamma9_n'+str(n_back_step)+'_k5_'+
'_lr'+str(learning_rate)+
'_threshold'+str(eps_threshold)+
'_eps'+str(epsilon_greedy_init)+
'_penalty_reward_scalar_'+str(me.penalty_reward_scalar)+
'_use_experimental_metab_' +str(int(use_experimental_data))+
'_sim'+str(sim_number)+
'.txt', epr_per_state, fmt='%f')