def test_forfatal_from_dict(self):
"""
Test if dictionary-based constraint interface is working.
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
np.random.seed(1)
n_cells = 2000
n_genes = 2
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
# Build design matrix:
sample_description = pd.DataFrame({
"cond": ["cond" + str(i // 1000) for i in range(n_cells)],
"batch": ["batch" + str(i // 500) for i in range(n_cells)]
})
test = de.test.wald(data=sim.input_data,
sample_description=sample_description,
formula_loc="~1+cond+batch",
formula_scale="~1+cond+batch",
constraints_loc={"batch": "cond"},
constraints_scale={"batch": "cond"},
coef_to_test=["cond[T.cond1]"])
_ = test.summary()
def test_rank_test_zero_variance(self):
"""
Test if rank test works if it is given genes with zero variance.
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
np.random.seed(1)
sim = Simulator(num_observations=1000, num_features=10)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
sim.input_data.x[:, 0] = 0
sim.input_data.x[:, 1] = 5
random_sample_description = pd.DataFrame(
{"condition": np.random.randint(2, size=sim.nobs)})
test = de.test.rank_test(data=sim.input_data,
sample_description=random_sample_description,
grouping="condition",
is_sig_zerovar=True)
assert np.isnan(test.pval[0]) and test.pval[1] == 1, \
"rank test did not assign p-value of zero to groups with zero variance and same mean, %f, %f" % \
(test.pval[0], test.pval[1])
return True
def _test_null_distribution_rank(self, n_cells: int, n_genes: int):
"""
Test if de.test.rank_test() generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
"""
from batchglm.api.models.tf1.glm_norm import Simulator
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
random_sample_description = pd.DataFrame(
{"condition": np.random.randint(2, size=sim.nobs)})
test = de.test.rank_test(data=sim.input_data,
sample_description=random_sample_description,
grouping="condition")
_ = test.summary()
# Compare p-value distribution under null model against uniform distribution.
pval_h0 = stats.kstest(test.pval, 'uniform').pvalue
logging.getLogger("diffxpy").info(
'KS-test pvalue for null model match of rank_test(): %f' % pval_h0)
assert pval_h0 > 0.05, ("KS-Test failed: pval_h0=%f is <= 0.05!" %
np.round(pval_h0, 5))
return True
def simulate(self, n_cells: int = 200, n_genes: int = 2):
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
random_sample_description = pd.DataFrame({
"condition": np.random.randint(2, size=sim.input_data.num_observations)
})
return sim.x, random_sample_description
def test_forfatal_from_string(self):
"""
Test if _from_string interface is working.
n_cells is constant as the design matrix and constraints depend on it.
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
np.random.seed(1)
n_cells = 2000
n_genes = 2
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
# Build design matrix:
dmat = np.zeros([n_cells, 6])
dmat[:, 0] = 1
dmat[:500, 1] = 1 # bio rep 1
dmat[500:1000, 2] = 1 # bio rep 2
dmat[1000:1500, 3] = 1 # bio rep 3
dmat[1500:2000, 4] = 1 # bio rep 4
dmat[1000:2000, 5] = 1 # condition effect
coefficient_names = [
'intercept', 'bio1', 'bio2', 'bio3', 'bio4', 'treatment1'
]
dmat_est = pd.DataFrame(data=dmat, columns=coefficient_names)
dmat_est_loc, _ = de.utils.design_matrix(dmat=dmat_est,
return_type="dataframe")
dmat_est_scale, _ = de.utils.design_matrix(dmat=dmat_est,
return_type="dataframe")
# Build constraints:
constraints_loc = de.utils.constraint_matrix_from_string(
dmat=dmat_est_loc.values,
coef_names=dmat_est_loc.columns,
constraints=["bio1+bio2=0", "bio3+bio4=0"])
constraints_scale = de.utils.constraint_matrix_from_string(
dmat=dmat_est_scale.values,
coef_names=dmat_est_scale.columns,
constraints=["bio1+bio2=0", "bio3+bio4=0"])
test = de.test.wald(data=sim.input_data,
dmat_loc=dmat_est_loc,
dmat_scale=dmat_est_scale,
constraints_loc=constraints_loc,
constraints_scale=constraints_scale,
coef_to_test=["treatment1"])
_ = test.summary()
def test_null_distribution_lrt(self,
n_cells: int = 4000,
n_genes: int = 200):
"""
Test if de.lrt() generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=2)
sim.generate()
sample_description = pd.DataFrame({
"covar1":
np.random.randint(2, size=sim.nobs),
"covar2":
np.random.randint(2, size=sim.nobs)
})
sample_description["cond"] = sim.sample_description["condition"].values
partition = de.test.partition(data=sim.x,
parts="cond",
sample_description=sample_description)
det = partition.lrt(full_formula_loc="~ 1 + covar1",
full_formula_scale="~ 1",
reduced_formula_loc="~ 1",
reduced_formula_scale="~ 1",
training_strategy="DEFAULT",
dtype="float64")
_ = det.summary()
# Compare p-value distribution under null model against uniform distribution.
pval_h0 = stats.kstest(det.pval.flatten(), 'uniform').pvalue
logging.getLogger("diffxpy").info(
'KS-test pvalue for null model match of lrt(): %f' % pval_h0)
assert pval_h0 > 0.05, "KS-Test failed: pval_h0=%f is <= 0.05!" % np.round(
pval_h0, 5)
return True
def test_null_distribution_wald(self,
n_cells: int = 2000,
n_genes: int = 100,
n_groups: int = 2):
"""
Test if de.test_wald_loc() generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distriubution and a uniform distribution.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
from batchglm.api.models.tf1.glm_nb import Simulator
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
random_sample_description = pd.DataFrame(
{"condition": np.random.randint(n_groups, size=sim.nobs)})
test = de.test.versus_rest(
data=sim.x,
grouping="condition",
test="wald",
noise_model="nb",
sample_description=random_sample_description,
batch_size=500,
training_strategy="DEFAULT",
dtype="float64")
summary = test.summary()
# Compare p-value distribution under null model against uniform distribution.
pval_h0 = stats.kstest(test.pval.flatten(), 'uniform').pvalue
logging.getLogger("diffxpy").info(
'KS-test pvalue for null model match of test_wald_loc(): %f' %
pval_h0)
assert pval_h0 > 0.05, "KS-Test failed: pval_h0=%f is <= 0.05!" % np.round(
pval_h0, 5)
return True
def test_null_distribution_wald_constrained(self, n_genes: int = 100):
"""
Test if de.wald() with constraints generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
n_cells is constant as the design matrix and constraints depend on it.
:param n_genes: Number of genes to simulate (number of tests).
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
np.random.seed(1)
n_cells = 2000
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
# Build design matrix:
sample_description = pd.DataFrame({
"cond": ["cond" + str(i // 1000) for i in range(n_cells)],
"batch": ["batch" + str(i // 500) for i in range(n_cells)]
})
test = de.test.wald(data=sim.input_data,
sample_description=sample_description,
formula_loc="~1+cond+batch",
formula_scale="~1+cond+batch",
constraints_loc={"batch": "cond"},
constraints_scale={"batch": "cond"},
coef_to_test=["cond[T.cond1]"])
_ = test.summary()
# Compare p-value distribution under null model against uniform distribution.
pval_h0 = stats.kstest(test.pval, 'uniform').pvalue
logging.getLogger("diffxpy").info(
'KS-test pvalue for null model match of wald(): %f' % pval_h0)
assert pval_h0 > 0.05, "KS-Test failed: pval_h0 is <= 0.05!"
return True
def _test_null_distribution_lrt(self, n_cells: int, n_genes: int,
noise_model: str):
"""
Test if de.lrt() generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
:param noise_model: Noise model to use for data fitting.
"""
if noise_model == "nb":
from batchglm.api.models.tf1.glm_nb import Simulator
elif noise_model == "norm":
from batchglm.api.models.tf1.glm_norm import Simulator
else:
raise ValueError("noise model %s not recognized" % noise_model)
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
random_sample_description = pd.DataFrame(
{"condition": np.random.randint(2, size=sim.nobs)})
test = de.test.lrt(data=sim.input_data,
sample_description=random_sample_description,
full_formula_loc="~ 1 + condition",
full_formula_scale="~ 1",
reduced_formula_loc="~ 1",
reduced_formula_scale="~ 1",
noise_model=noise_model)
_ = test.summary()
# Compare p-value distribution under null model against uniform distribution.
pval_h0 = stats.kstest(test.pval, 'uniform').pvalue
logging.getLogger("diffxpy").info(
'KS-test pvalue for null model match of lrt(): %f' % pval_h0)
assert pval_h0 > 0.05, ("KS-Test failed: pval_h0=%f is <= 0.05!" %
np.round(pval_h0, 5))
return True
def simulate(self):
if self.noise_model is None:
raise ValueError("noise_model is None")
else:
if self.noise_model == "nb":
from batchglm.api.models.tf1.glm_nb import Simulator
elif self.noise_model == "norm":
from batchglm.api.models import Simulator
elif self.noise_model == "beta":
from batchglm.api.models.tf1.glm_beta import Simulator
else:
raise ValueError("noise_model not recognized")
num_observations = 500
sim = Simulator(num_observations=num_observations, num_features=4)
sim.generate_sample_description(num_conditions=2, num_batches=2)
sim.generate()
self.sim = sim
def _test_residuals_fit(
self,
n_cells: int,
n_genes: int,
noise_model: str
):
"""
Test if de.wald() (multivariate mode) generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
:param n_cells: Number of cells to simulate (number of observations per test).
:param n_genes: Number of genes to simulate (number of tests).
:param noise_model: Noise model to use for data fitting.
"""
if noise_model == "nb":
from batchglm.api.models.tf1.glm_nb import Simulator
elif noise_model == "norm":
from batchglm.api.models.tf1.glm_norm import Simulator
else:
raise ValueError("noise model %s not recognized" % noise_model)
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
random_sample_description = pd.DataFrame({
"condition": np.random.randint(2, size=sim.nobs),
"batch": np.random.randint(2, size=sim.nobs)
})
res = de.fit.residuals(
data=sim.input_data,
sample_description=random_sample_description,
formula_loc="~ 1 + condition + batch",
noise_model=noise_model
)
return True
def test_for_fatal(self):
"""
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
sim = Simulator(num_observations=50, num_features=10)
sim.generate_sample_description(num_batches=0, num_conditions=2)
sim.generate()
test = de.test.wald(data=sim.X,
factor_loc_totest="condition",
formula_loc="~ 1 + condition",
sample_description=sim.sample_description,
gene_names=[str(x) for x in range(sim.X.shape[1])],
training_strategy="DEFAULT",
dtype="float64")
# Set up reference gene sets.
rs = de.enrich.RefSets()
rs.add(id="set1", source="manual", gene_ids=["1", "3"])
rs.add(id="set2", source="manual", gene_ids=["5", "6"])
for i in [True, False]:
for j in [True, False]:
enrich_test_i = de.enrich.test(
ref=rs,
det=test,
threshold=0.05,
incl_all_zero=i,
clean_ref=j,
)
_ = enrich_test_i.summary()
_ = enrich_test_i.significant_set_ids()
_ = enrich_test_i.significant_sets()
_ = enrich_test_i.set_summary(id="set1")
return True
def _test_null_distribution_wald_constrained_2layer(
self, n_genes: int = 100):
"""
Test if de.wald() with constraints generates a uniform p-value distribution
if it is given data simulated based on the null model. Returns the p-value
of the two-side Kolmgorov-Smirnov test for equality of the observed
p-value distribution and a uniform distribution.
n_cells is constant as the design matrix and constraints depend on it.
:param n_genes: Number of genes to simulate (number of tests).
"""
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("batchglm").setLevel(logging.WARNING)
logging.getLogger("diffxpy").setLevel(logging.WARNING)
np.random.seed(1)
n_cells = 12000
sim = Simulator(num_observations=n_cells, num_features=n_genes)
sim.generate_sample_description(num_batches=0, num_conditions=0)
sim.generate()
# Build design matrix:
dmat = np.zeros([n_cells, 14])
dmat[:, 0] = 1
dmat[6000:12000, 1] = 1 # condition effect
dmat[:1000, 2] = 1 # bio rep 1 - treated 1
dmat[1000:3000, 3] = 1 # bio rep 2 - treated 2
dmat[3000:5000, 4] = 1 # bio rep 3 - treated 3
dmat[5000:6000, 5] = 1 # bio rep 4 - treated 4
dmat[6000:7000, 6] = 1 # bio rep 5 - untreated 1
dmat[7000:9000, 7] = 1 # bio rep 6 - untreated 2
dmat[9000:11000, 8] = 1 # bio rep 7 - untreated 3
dmat[11000:12000, 9] = 1 # bio rep 8 - untreated 4
dmat[1000:2000, 10] = 1 # tech rep 1
dmat[7000:8000, 10] = 1 # tech rep 1
dmat[2000:3000, 11] = 1 # tech rep 2
dmat[8000:9000, 11] = 1 # tech rep 2
dmat[3000:4000, 12] = 1 # tech rep 3
dmat[9000:10000, 12] = 1 # tech rep 3
dmat[4000:5000, 13] = 1 # tech rep 4
dmat[10000:11000, 13] = 1 # tech rep 4
coefficient_names = [
'intercept', 'treatment1', 'bio1', 'bio2', 'bio3', 'bio4', 'bio5',
'bio6', 'bio7', 'bio8', 'tech1', 'tech2', 'tech3', 'tech4'
]
dmat_est = pd.DataFrame(data=dmat, columns=coefficient_names)
dmat_est_loc = de.utils.design_matrix(dmat=dmat_est,
return_type="dataframe")
dmat_est_scale = de.utils.design_matrix(dmat=dmat_est.iloc[:, [0]],
return_type="dataframe")
# Build constraints:
constraints_loc = de.utils.constraint_matrix_from_string(
dmat=dmat_est_loc.values,
coef_names=dmat_est_loc.columns,
constraints=[
"bio1+bio2=0", "bio3+bio4=0", "bio5+bio6=0", "bio7+bio8=0",
"tech1+tech2=0", "tech3+tech4=0"
])
constraints_scale = None
test = de.test.wald(data=sim.input_data,
dmat_loc=dmat_est_loc,
dmat_scale=dmat_est_scale,
constraints_loc=constraints_loc,
constraints_scale=constraints_scale,
coef_to_test=["treatment1"])
_ = test.summary()
# Compare p-value distribution under null model against uniform distribution.
pval_h0 = stats.kstest(test.pval, 'uniform').pvalue
logging.getLogger("diffxpy").info(
'KS-test pvalue for null model match of wald(): %f' % pval_h0)
assert pval_h0 > 0.05, "KS-Test failed: pval_h0 is <= 0.05!"
return True
def _test_compute_hessians(self, sparse):
if self.noise_model is None:
raise ValueError("noise_model is None")
else:
if self.noise_model == "nb":
from batchglm.api.models.tf1.glm_nb import Simulator, InputDataGLM
elif self.noise_model == "norm":
from batchglm.api.models import Simulator, InputDataGLM
elif self.noise_model == "beta":
from batchglm.api.models.tf1.glm_beta import Simulator, InputDataGLM
else:
raise ValueError("noise_model not recognized")
num_observations = 500
num_conditions = 2
sim = Simulator(num_observations=num_observations, num_features=4)
sim.generate_sample_description(num_conditions=num_conditions,
num_batches=2)
sim.generate()
sample_description = data_utils.sample_description_from_xarray(
sim.data, dim="observations")
design_loc = data_utils.design_matrix(
sample_description, formula="~ 1 + condition + batch")
design_scale = data_utils.design_matrix(sample_description,
formula="~ 1 + condition")
if sparse:
input_data = InputDataGLM(data=scipy.sparse.csr_matrix(sim.X),
design_loc=design_loc,
design_scale=design_scale)
else:
input_data = InputDataGLM(data=sim.X,
design_loc=design_loc,
design_scale=design_scale)
# Compute hessian based on analytic solution.
pkg_constants.HESSIAN_MODE = "analytic"
t0_analytic = time.time()
h_analytic = self.get_hessians(input_data)
t1_analytic = time.time()
t_analytic = t1_analytic - t0_analytic
# Compute hessian based on tensorflow auto-differentiation.
pkg_constants.HESSIAN_MODE = "tf1"
t0_tf = time.time()
h_tf = self.get_hessians(input_data)
t1_tf = time.time()
t_tf = t1_tf - t0_tf
logging.getLogger("batchglm").info(
"run time observation batch-wise analytic solution: %f" %
t_analytic)
logging.getLogger("batchglm").info("run time tensorflow solution: %f" %
t_tf)
logging.getLogger("batchglm").info("MAD: %f" %
np.max(np.abs((h_tf - h_analytic))))
#i = 1
#print(h_tf[i, :, :])
#print(h_analytic[i, :, :])
#print(h_tf[i, :, :] - h_analytic[i, :, :])
# Make sure that hessians are not all zero which might make evaluation of equality difficult.
assert np.sum(np.abs(h_analytic)) > 1e-10, \
"hessians too small to perform test: %f" % np.sum(np.abs(h_analytic))
mad = np.max(np.abs(h_tf - h_analytic))
assert mad < 1e-15, mad
return True