def test_global_fit(self):
#kf = KineticsFitting(self.series_apo, bounds=(1e-2, 800), temperature=self.temperature, pH=self.pH)
initial_rates = csv_to_protein(
os.path.join(directory, 'test_data', 'ecSecB_guess.txt'))
t0 = time.time() # Very crude benchmarks
gibbs_guess = self.series_apo.guess_deltaG(
initial_rates['rate']).to_numpy()
fr_global = fit_gibbs_global(self.series_apo,
gibbs_guess,
epochs=1000,
r1=2)
t1 = time.time()
assert t1 - t0 < 5
out_deltaG = fr_global.output
check_deltaG = csv_to_protein(
os.path.join(directory, 'test_data', 'ecSecB_torch_fit.txt'))
assert np.allclose(check_deltaG['deltaG'],
out_deltaG['deltaG'],
equal_nan=True,
rtol=0.01)
assert np.allclose(check_deltaG['covariance'],
out_deltaG['covariance'],
equal_nan=True,
rtol=0.01)
assert np.allclose(check_deltaG['k_obs'],
out_deltaG['k_obs'],
equal_nan=True,
rtol=0.01)
def test_dtype_cuda(self):
check_deltaG = csv_to_protein(output_dir / 'ecSecB_torch_fit.csv')
initial_rates = csv_to_dataframe(output_dir / 'ecSecB_guess.csv')
cfg.set('fitting', 'device', 'cuda')
gibbs_guess = self.hdxm_apo.guess_deltaG(
initial_rates['rate']).to_numpy()
if torch.cuda.is_available():
fr_global = fit_gibbs_global(self.hdxm_apo,
gibbs_guess,
epochs=1000,
r1=2)
out_deltaG = fr_global.output
for field in ['dG', 'k_obs', 'covariance']:
assert_series_equal(check_deltaG[field],
out_deltaG[self.hdxm_apo.name, field],
rtol=0.01,
check_dtype=False,
check_names=False)
else:
with pytest.raises(AssertionError, match=r".* CUDA .*"):
fr_global = fit_gibbs_global(self.hdxm_apo,
gibbs_guess,
epochs=1000,
r1=2)
cfg.set('fitting', 'device', 'cpu')
cfg.set('fitting', 'dtype', 'float32')
fr_global = fit_gibbs_global(self.hdxm_apo,
gibbs_guess,
epochs=1000,
r1=2)
dg = fr_global.model.dG
assert dg.dtype == torch.float32
out_deltaG = fr_global.output
for field in ['dG', 'k_obs']:
assert_series_equal(check_deltaG[field],
out_deltaG[self.hdxm_apo.name, field],
rtol=0.01,
check_dtype=False,
check_names=False)
cfg.set('fitting', 'dtype', 'float64')
def setup_class(cls):
cls.fpath = input_dir / 'ecSecB_apo.csv'
data = read_dynamx(cls.fpath)
control = ('Full deuteration control', 0.167*60)
cls.temperature, cls.pH = 273.15 + 30, 8.
pf = PeptideMasterTable(data, drop_first=1, ignore_prolines=True, remove_nan=False)
pf.set_control(control)
cls.hdxm = HDXMeasurement(pf.get_state('SecB WT apo'), temperature=cls.temperature, pH=cls.pH)
initial_rates = csv_to_dataframe(output_dir / 'ecSecB_guess.csv')
gibbs_guess = cls.hdxm.guess_deltaG(initial_rates['rate'])
cls.fit_result = fit_gibbs_global(cls.hdxm, gibbs_guess, epochs=100, r1=2)
def test_global_fit_extended(self):
check_deltaG = csv_to_protein(output_dir /
'ecSecB_torch_fit_epochs_20000.csv')
initial_rates = csv_to_dataframe(output_dir / 'ecSecB_guess.csv')
gibbs_guess = self.hdxm_apo.guess_deltaG(initial_rates['rate'])
t0 = time.time() # Very crude benchmarks
fr_global = fit_gibbs_global(self.hdxm_apo,
gibbs_guess,
epochs=20000,
r1=2)
t1 = time.time()
assert t1 - t0 < 50
out_deltaG = fr_global.output
for field in ['dG', 'k_obs', 'covariance']:
assert_series_equal(check_deltaG[self.hdxm_apo.name, field],
out_deltaG[self.hdxm_apo.name, field],
rtol=0.01,
check_dtype=False)
errors = fr_global.get_squared_errors()
assert errors.shape == (1, self.hdxm_apo.Np, self.hdxm_apo.Nt)
def test_global_fit_extended_cuda(self):
check_deltaG = csv_to_protein(output_dir /
'ecSecB_torch_fit_epochs_20000.csv')
initial_rates = csv_to_dataframe(output_dir / 'ecSecB_guess.csv')
gibbs_guess = self.hdxm_apo.guess_deltaG(initial_rates['rate'])
# todo allow contextmanger?
cfg.set('fitting', 'device', 'cuda')
cfg.set('fitting', 'dtype', 'float32')
fr_global = fit_gibbs_global(self.hdxm_apo,
gibbs_guess,
epochs=20000,
r1=2)
out_deltaG = fr_global.output
for field in ['dG', 'k_obs']:
assert_series_equal(check_deltaG[self.hdxm_apo.name, field],
out_deltaG[self.hdxm_apo.name, field],
rtol=0.01,
check_dtype=False)
cfg.set('fitting', 'device', 'cpu')
cfg.set('fitting', 'dtype', 'float64')
pmt = PeptideMasterTable(data, drop_first=1, ignore_prolines=True, remove_nan=False)
pmt.set_control(control)
temperature, pH = 273.15 + 30, 8.
hdxm = HDXMeasurement(pmt.get_state('SecB WT apo'), sequence=sequence, temperature=temperature, pH=pH)
if guess:
client = default_client()
wt_avg_result = fit_rates_weighted_average(hdxm, bounds=(1e-2, 800))
output = wt_avg_result.output
output.to_file(directory / 'test_data' / 'ecSecB_guess.txt')
else:
output = csv_to_protein(directory / 'test_data' / 'ecSecB_guess.txt')
gibbs_guess = hdxm.guess_deltaG(output['rate'])
fr_torch = fit_gibbs_global(hdxm, gibbs_guess, epochs=epochs, r1=2)
fr_torch.output.to_file(directory / 'test_data' / 'ecSecB_torch_fit.txt')
hdxm_dimer = HDXMeasurement(pmt.get_state('SecB his dimer apo'), sequence=sequence_dimer,
temperature=temperature, pH=pH)
hdx_set = HDXMeasurementSet([hdxm_dimer, hdxm])
gibbs_guess = hdx_set.guess_deltaG([output['rate'], output['rate']])
batch_result = fit_gibbs_global_batch(hdx_set, gibbs_guess, epochs=epochs)
batch_result.output.to_file(directory / 'test_data' / 'ecSecB_batch.csv')
batch_result.output.to_file(directory / 'test_data' / 'ecSecB_batch.txt', fmt='pprint')
# Order is inverted compared to test!
mock_alignment = {
temperature=temperature,
pH=pH)
data = pmt.get_state('SecB WT apo')
reduced_data = data[data['end'] < 40]
hdxm_reduced = HDXMeasurement(reduced_data, temperature=temperature, pH=pH)
result = fit_rates_weighted_average(hdxm_reduced)
reduced_guess = result.output
dataframe_to_file(output_dir / 'ecSecB_reduced_guess.csv', reduced_guess)
dataframe_to_file(output_dir / 'ecSecB_reduced_guess.txt',
reduced_guess,
fmt='pprint')
gibbs_guess = hdxm_reduced.guess_deltaG(reduced_guess['rate'])
fr_torch = fit_gibbs_global(hdxm_reduced, gibbs_guess, epochs=epochs, r1=2)
save_fitresult(output_dir / 'ecsecb_reduced', fr_torch)
if guess:
wt_avg_result = fit_rates_weighted_average(hdxm,
bounds=(1e-2 / 60., 800 / 60.))
guess_output = wt_avg_result.output
dataframe_to_file(output_dir / 'ecSecB_guess.csv', guess_output)
dataframe_to_file(output_dir / 'ecSecB_guess.txt',
guess_output,
fmt='pprint')
else:
guess_output = csv_to_dataframe(output_dir / 'ecSecB_guess.csv')
# Export protein sequence and intrinsic rate of exchange
hdxm.coverage.protein.to_file(output_dir / 'ecSecB_info.csv')
# Load the data of two Dynamx files, and combine the result to one table
data = read_dynamx(input_dir / 'ecSecB_apo.csv', input_dir / 'ecSecB_dimer.csv')
pmt = PeptideMasterTable(data, drop_first=1, ignore_prolines=True, remove_nan=False)
pmt.set_control(('Full deuteration control', 0.167*60))
temperature, pH = 273.15 + 30, 8.
hdxm = HDXMeasurement(pmt.get_state('SecB WT apo'), temperature=temperature, pH=pH)
#%%
if guess:
client = default_client()
wt_avg_result = fit_rates_weighted_average(hdxm, client=client)
init_guess = wt_avg_result.output
else:
init_guess = csv_to_dataframe(test_data_dir / 'output' / 'ecSecB_guess.csv')
gibbs_guess = hdxm.guess_deltaG(init_guess['rate'])
#%%
fr_torch = fit_gibbs_global(hdxm, gibbs_guess, **fit_kwargs)
#Human readable output
fr_torch.to_file(output_dir / 'SecB_fit_result.txt', fmt='pprint')
#Machine readable output
fr_torch.to_file(output_dir / 'SecB_fit_result.csv', fmt='csv')
save_fitresult(output_dir / 'SecB_fit', fr_torch)
guess = False
epochs = 1000
root_dir = Path().resolve().parent
test_data_dir = root_dir / 'tests' / 'test_data'
input_file_path = test_data_dir / 'ecSecB_apo.csv'
# Load the data of two Dynamx files, and combine the result to one table
data = read_dynamx(test_data_dir / 'ecSecB_apo.csv',
test_data_dir / 'ecSecB_dimer.csv')
pmt = PeptideMasterTable(data,
drop_first=1,
ignore_prolines=True,
remove_nan=False)
pmt.set_control(('Full deuteration control', 0.167))
temperature, pH = 273.15 + 30, 8.
series = HDXMeasurement(pmt.get_state('SecB WT apo'),
temperature=temperature,
pH=pH)
if guess:
client = default_client()
wt_avg_result = fit_rates_weighted_average(series, client=client)
init_guess = wt_avg_result.output
else:
init_guess = csv_to_protein(test_data_dir / 'ecSecB_guess.txt')
gibbs_guess = series.guess_deltaG(init_guess['rate'])
fr_torch = fit_gibbs_global(series, gibbs_guess, epochs=epochs)
print(fr_torch.metadata['total_loss'])