def test_filter_genes():
"""
Ensure genes with no expression are pre-filtered
"""
# Simulate some data
N_CELLS = 100
N_DIM = 10
N_GENES = 10
N_GENES_ZERO = 5
latent = sim_data.sim_latent(N_CELLS, N_DIM)
latent = pd.DataFrame(latent)
umi_counts = sim_data.sim_umi_counts(N_CELLS, 2000, 200)
umi_counts = pd.Series(umi_counts)
gene_exp = np.random.rand(N_GENES + N_GENES_ZERO, N_CELLS)
gene_exp[N_GENES:] = 0
gene_exp = pd.DataFrame(
gene_exp,
index=['Gene{}'.format(i + 1) for i in range(gene_exp.shape[0])],
columns=latent.index)
hs = Hotspot(gene_exp,
model='normal',
latent=latent,
umi_counts=umi_counts)
assert hs.counts.shape[0] == N_GENES
def test_local_correlation():
N_CELLS = 1000
N_DIM = 10
latent = sim_data.sim_latent(N_CELLS, N_DIM)
latent = pd.DataFrame(latent)
umi_counts = sim_data.sim_umi_counts(N_CELLS, 2000, 200)
umi_counts = pd.Series(umi_counts)
neighbors, weights = neighbors_and_weights(latent,
n_neighbors=30,
neighborhood_factor=3)
neighbors = neighbors.values
weights = weights.values
weights = make_weights_non_redundant(neighbors, weights)
counts_i = np.random.randn(N_CELLS)
gxy = local_cov_pair(counts_i, counts_i, neighbors, weights)
g = local_cov_weights(counts_i, neighbors, weights)
assert math.isclose(
g, gxy, rel_tol=1e-10
), "Pairwise covariance on (x, x) should be same as local covariance on (x)"
def test_local_autocorrelation_centered():
"""
Test if the expected moment calculation is correct
"""
# Simulate some data
N_CELLS = 1000
N_DIM = 10
latent = sim_data.sim_latent(N_CELLS, N_DIM)
latent = pd.DataFrame(latent)
umi_counts = sim_data.sim_umi_counts(N_CELLS, 2000, 200)
umi_counts = pd.Series(umi_counts)
neighbors, weights = neighbors_and_weights(latent,
n_neighbors=30,
neighborhood_factor=3)
neighbors = neighbors.values
weights = weights.values
weights = make_weights_non_redundant(neighbors, weights)
Wtot2 = (weights**2).sum()
# Simulate counts
for gene_p in [.2, 1, 5, 10]:
N_REPS = 10000
mu, var, x2 = bernoulli_model.true_params_scaled(
gene_p, umi_counts.values)
Gs = []
for i in range(N_REPS):
counts_i = sim_data.sim_counts_bernoulli(N_CELLS,
umi_counts.values, gene_p)
counts_i = center_values(counts_i, mu, var)
g = local_stats.local_cov_weights(counts_i, neighbors, weights)
Gs.append(g)
Gs = np.array(Gs)
EG = 0
EG2 = Wtot2
EstdG = (EG2 - EG**2)**0.5
Gmean = Gs.mean()
Gstd = Gs.std()
assert math.isclose(
0, abs(Gmean / Gstd), abs_tol=5e-2
), "EG is off for gene_p={}, Actual={:.2f}, Expected={:.2f}".format(
gene_p, Gmean, EG)
assert math.isclose(
EstdG, Gstd, rel_tol=5e-2
), "stdG is off for gene_p={}, Actual={:.2f}, Expected={:.2f}".format(
gene_p, Gstd, EstdG)
def test_models():
"""
Ensure each model runs
"""
# Simulate some data
N_CELLS = 100
N_DIM = 10
N_GENES = 10
latent = sim_data.sim_latent(N_CELLS, N_DIM)
latent = pd.DataFrame(
latent, index=['Cell{}'.format(i + 1) for i in range(N_CELLS)])
umi_counts = sim_data.sim_umi_counts(N_CELLS, 2000, 200)
umi_counts = pd.Series(umi_counts)
gene_exp = np.random.rand(N_GENES, N_CELLS)
gene_exp = pd.DataFrame(
gene_exp,
index=['Gene{}'.format(i + 1) for i in range(gene_exp.shape[0])],
columns=latent.index)
for model in ['danb', 'bernoulli', 'normal', 'none']:
hs = Hotspot(gene_exp,
model=model,
latent=latent,
umi_counts=umi_counts)
hs.create_knn_graph(False, n_neighbors=30)
hs.compute_hotspot()
assert isinstance(hs.results, pd.DataFrame)
assert hs.results.shape[0] == N_GENES
hs.compute_autocorrelations()
assert isinstance(hs.results, pd.DataFrame)
assert hs.results.shape[0] == N_GENES
hs.compute_local_correlations(gene_exp.index)
assert isinstance(hs.local_correlation_z, pd.DataFrame)
assert hs.local_correlation_z.shape[0] == N_GENES
assert hs.local_correlation_z.shape[1] == N_GENES
hs.create_modules(min_gene_threshold=2, fdr_threshold=1)
assert isinstance(hs.modules, pd.Series)
assert (hs.modules.index & gene_exp.index).size == N_GENES
assert isinstance(hs.linkage, np.ndarray)
assert hs.linkage.shape == (N_GENES - 1, 4)
hs.calculate_module_scores()
assert isinstance(hs.module_scores, pd.DataFrame)
assert (hs.module_scores.index == gene_exp.columns).all()