def simulate_pcrs_with_probabilities():
num_experiments = 100
probabilities = [0.01 * i for i in range(1, 10)]
num_pcrs = 5
y = []
naive_y = []
for probability in probabilities:
state_probabilities = {
'contaminated': probability,
'defective': probability,
'good': 1 - probability
}
errors = []
naive_errors = []
for i in range(num_experiments):
db = PCRDatabase()
logic = PCRLogic(db)
curr_aliquots = [
Aliquot(reagent, 1, 'Biowares', choose(state_probabilities))
for reagent in REAGENT_MAP
]
for j in range(num_pcrs):
false_neg = any(
map(lambda aliquot: aliquot.state == 'defective',
curr_aliquots))
false_pos = any(
map(lambda aliquot: aliquot.state == 'contaminated',
curr_aliquots)) and not false_neg
if false_neg:
pcr = PCR(False, False, curr_aliquots)
elif false_pos:
pcr = PCR(True, True, curr_aliquots)
else:
pcr = PCR(True, False, curr_aliquots)
db.add_pcr(pcr)
# Randomly choose new aliquot
for i in range(len(curr_aliquots)):
old_aliquot = curr_aliquots[i]
new_aliquot = Aliquot(old_aliquot.reagent,
old_aliquot.id + 1,
old_aliquot.manufacturer,
choose(state_probabilities))
aliquot_probabilities = {
old_aliquot: 0.8,
new_aliquot: 0.2
}
curr_aliquots[i] = choose(aliquot_probabilities)
# Compute the defective and contaminated aliquot probabilities
deductions = logic.make_probabilistic_deductions()
all_aliquots = db.get_all_aliquots()
errors.append(rmse(all_aliquots, deductions))
# Compute naive approach errors
naive_deductions = dict(deductions)
for error in naive_deductions:
prob_list = naive_deductions[error]
for i in range(len(prob_list)):
prob_list[i] = (prob_list[i][0], 1.0)
naive_errors.append(rmse(all_aliquots, naive_deductions))
print[(pcr.pos_control_result, pcr.negative_control_result)
for pcr in db.pcrs]
avg_error = sum(errors) / len(errors)
avg_naive_error = sum(naive_errors) / len(naive_errors)
y.append(avg_error)
naive_y.append(avg_naive_error)
print y
print naive_y
pyplot.plot(probabilities, y, label='Bayesian inference')
pyplot.plot(probabilities, naive_y, label='Naive approach')
pyplot.legend(loc=2)
pyplot.xlabel('Probability of reagent being defective or contaminated')
pyplot.ylabel('RMSE')
pyplot.title(
'RMSE as a function of reagent defective/contaminated probability')
pyplot.show()
def simulate_pcrs_with_heatmap():
num_pcrs = 10
state_probabilities = {
'contaminated': 0.04,
'defective': 0.04,
'good': 0.92
}
db = PCRDatabase()
logic = PCRLogic(db)
curr_aliquots = [
Aliquot(reagent, 1, 'Biowares', choose(state_probabilities))
for reagent in REAGENT_MAP
]
for j in range(num_pcrs):
false_neg = any(
map(lambda aliquot: aliquot.state == 'defective', curr_aliquots))
false_pos = any(
map(lambda aliquot: aliquot.state == 'contaminated',
curr_aliquots)) and not false_neg
if false_neg:
pcr = PCR(False, False, curr_aliquots)
elif false_pos:
pcr = PCR(True, True, curr_aliquots)
else:
pcr = PCR(True, False, curr_aliquots)
db.add_pcr(pcr)
# Randomly choose new aliquot
for i in range(len(curr_aliquots)):
old_aliquot = curr_aliquots[i]
new_aliquot = Aliquot(old_aliquot.reagent, old_aliquot.id + 1,
old_aliquot.manufacturer,
choose(state_probabilities))
aliquot_probabilities = {old_aliquot: 0.8, new_aliquot: 0.2}
curr_aliquots[i] = choose(aliquot_probabilities)
# Compute the defective and contaminated aliquot probabilities
deductions = logic.make_probabilistic_deductions()
all_aliquots = db.get_all_aliquots()
print 'defective', dict(deductions['defective'])
print 'contaminated', dict(deductions['contaminated'])
# Create heatmap
actual_reagent_contaminations = np.array(
[[0. for _ in range(len(curr_aliquots))] for _ in range(num_pcrs)])
actual_reagent_defective = np.array(
[[0. for _ in range(len(curr_aliquots))] for _ in range(num_pcrs)])
predicted_reagent_contaminations = np.array(
[[0. for _ in range(len(curr_aliquots))] for _ in range(num_pcrs)])
predicted_reagent_defective = np.array(
[[0. for _ in range(len(curr_aliquots))] for _ in range(num_pcrs)])
for i in range(len(db.pcrs)):
pcr = db.pcrs[i]
for j in range(len(pcr.aliquots)):
aliquot = pcr.aliquots[j]
actual_reagent_contaminations[i][j] = float(
aliquot.state == 'contaminated')
actual_reagent_defective[i][j] = float(
aliquot.state == 'defective')
predicted_reagent_defective[i][j] = dict(
deductions['defective']).get(aliquot, 0.0)
predicted_reagent_contaminations[i][j] = dict(
deductions['contaminated']).get(aliquot, 0.0)
print[(pcr.pos_control_result, pcr.negative_control_result)
for pcr in db.pcrs]
# Plot heatmap for actual contaminations
row_labels = [REAGENT_MAP[aliquot.reagent] for aliquot in curr_aliquots]
column_labels = [i + 1 for i in range(num_pcrs)]
plot_heatmap(actual_reagent_contaminations, row_labels, column_labels,
matplotlib.cm.Reds, 'Actual contaminated reagents')
plot_heatmap(predicted_reagent_contaminations, row_labels, column_labels,
matplotlib.cm.Reds,
'Bayesian probabilities for contaminated reagents')
plot_heatmap(actual_reagent_defective, row_labels, column_labels,
matplotlib.cm.Blues, 'Actual defective reagents')
plot_heatmap(predicted_reagent_defective, row_labels, column_labels,
matplotlib.cm.Blues,
'Bayesian probabilities for defective reagents')
pyplot.show()