def test__mean_interhomolog_counts_ambig():
lengths = np.array([10, 20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
mean_chrom_coords = np.array([[0, 3, 0], [-4, 0, 0], [0, -3, 0], [4, 0,
0]])
n = lengths.sum()
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
begin = end = 0
for i in range(len(lengths) * ploidy):
end += np.tile(lengths, ploidy)[i]
X_true[begin:end] -= X_true[begin:end].mean(axis=0)
X_true[begin:end] += mean_chrom_coords[i]
begin = end
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
nchrom = lengths.shape[0]
X_true_lowres = decrease_struct_res(X_true,
multiscale_factor=lengths.max(),
lengths=lengths)
dis_lowres = euclidean_distances(X_true_lowres)
dis_lowres[np.tril_indices(dis_lowres.shape[0])] = np.nan
np.fill_diagonal(dis_lowres[:nchrom, nchrom:], np.nan)
approx_interhomo_dis_ambig = np.nanmean(dis_lowres)
ambig_counts_raw = counts[:n, :n] + counts[n:, n:] + counts[:n, n:] + \
counts[n:, :n]
ambig_counts = _format_counts(counts=sparse.coo_matrix(ambig_counts_raw),
lengths=lengths,
ploidy=ploidy,
beta=beta)
bias = 0.1 + random_state.rand(n)
ambig_counts_biased = ambig_counts_raw * bias.reshape(-1, 1) * \
bias.reshape(-1, 1).T
ambig_counts_biased = _format_counts(
counts=sparse.coo_matrix(ambig_counts_biased),
lengths=lengths,
ploidy=ploidy,
beta=beta)
mhs_k_ambig = constraints._mean_interhomolog_counts(ambig_counts,
lengths=lengths)
mhs_k_ambig_biased = constraints._mean_interhomolog_counts(
ambig_counts_biased, lengths=lengths, bias=bias)
assert_allclose(mhs_k_ambig, mhs_k_ambig_biased)
assert_allclose(mhs_k_ambig**(1 / alpha),
approx_interhomo_dis_ambig,
rtol=1e-1)
def test__mean_interhomolog_counts_unambig():
lengths = np.array([10, 20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
mean_chrom_coords = np.array([[0, 3, 0], [-4, 0, 0], [0, -3, 0], [4, 0,
0]])
n = lengths.sum()
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
begin = end = 0
for i in range(len(lengths) * ploidy):
end += np.tile(lengths, ploidy)[i]
X_true[begin:end] -= X_true[begin:end].mean(axis=0)
X_true[begin:end] += mean_chrom_coords[i]
begin = end
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
true_interhomo_dis = constraints._inter_homolog_dis(X_true,
lengths=lengths)
bias = 0.1 + random_state.rand(n)
counts_biased = counts * np.tile(bias, 2).reshape(-1, 1) * \
np.tile(bias, 2).reshape(-1, 1).T
ua_counts = _format_counts(counts=sparse.coo_matrix(counts),
lengths=lengths,
ploidy=ploidy,
beta=beta)
ua_counts_biased = _format_counts(counts=sparse.coo_matrix(counts_biased),
lengths=lengths,
ploidy=ploidy,
beta=beta)
mhs_k_ua = constraints._mean_interhomolog_counts(ua_counts,
lengths=lengths)
mhs_k_ua_biased = constraints._mean_interhomolog_counts(ua_counts_biased,
lengths=lengths,
bias=bias)
assert_allclose(mhs_k_ua, mhs_k_ua_biased)
assert_allclose(mhs_k_ua**(1 / alpha), true_interhomo_dis, rtol=1e-2)
def test_bcc_constraint():
lengths = np.array([20])
ploidy = 2
alpha, beta = -3., 1.
n = lengths.sum()
X_true = np.concatenate([
np.arange(n * ploidy).reshape(-1, 1),
np.zeros((n * ploidy, 1)),
np.zeros((n * ploidy, 1))
],
axis=1)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta)
constraint = constraints.Constraints(counts,
lengths=lengths,
ploidy=ploidy,
multiscale_factor=1,
constraint_lambdas={'bcc': 1},
constraint_params=None)
constraint.check()
obj = constraint.apply(X_true)['obj_bcc']
assert obj < 1e-6
def test_estimate_alpha_beta_diploid_unambig():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha_true, beta_true = -3., 2.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta_true * dis**alpha_true
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta_true)
alpha, obj, converged, _ = estimate_alpha_beta.estimate_alpha(
X=X_true, counts=counts, alpha_init=alpha_true, lengths=lengths)
beta = list(
estimate_alpha_beta._estimate_beta(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths,
verbose=False).values())[0]
assert converged
assert obj < -1e4
assert_array_almost_equal(alpha_true, alpha, decimal=5)
assert_array_almost_equal(beta_true, beta, decimal=3)
def test_estimate_alpha_beta_diploid_combo():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha_true, beta_true_single = -3., 4.
ratio_ambig, ratio_pa, ratio_ua = [1 / 3] * 3
bias = None
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
poisson_intensity = dis**alpha_true
ambig_counts = ratio_ambig * beta_true_single * poisson_intensity
ambig_counts[np.isnan(ambig_counts) | np.isinf(ambig_counts)] = 0
ambig_counts = ambig_counts[:n, :n] + ambig_counts[
n:, n:] + ambig_counts[:n, n:] + ambig_counts[n:, :n]
ambig_counts = np.triu(ambig_counts, 1)
ambig_counts = sparse.coo_matrix(ambig_counts)
pa_counts = ratio_pa * beta_true_single * poisson_intensity
pa_counts[np.isnan(pa_counts) | np.isinf(pa_counts)] = 0
pa_counts = pa_counts[:, :n] + pa_counts[:, n:]
np.fill_diagonal(pa_counts[:n, :], 0)
np.fill_diagonal(pa_counts[n:, :], 0)
pa_counts = sparse.coo_matrix(pa_counts)
ua_counts = ratio_ua * beta_true_single * poisson_intensity
ua_counts[np.isnan(ua_counts) | np.isinf(ua_counts)] = 0
ua_counts = np.triu(ua_counts, 1)
ua_counts = sparse.coo_matrix(ua_counts)
counts_raw = [ambig_counts, pa_counts, ua_counts]
beta_true = np.array([ratio_ambig, ratio_pa, ratio_ua]) * beta_true_single
counts = _format_counts(counts=counts_raw,
lengths=lengths,
ploidy=ploidy,
beta=beta_true)
alpha, obj, converged, _ = estimate_alpha_beta.estimate_alpha(
X=X_true,
counts=counts,
alpha_init=alpha_true,
lengths=lengths,
bias=bias)
beta = list(
estimate_alpha_beta._estimate_beta(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths,
bias=bias,
verbose=False).values())[0]
assert converged
assert obj < -1e4
assert_array_almost_equal(alpha_true, alpha, decimal=5)
assert_array_almost_equal(beta_true, beta, decimal=3)
def test_poisson_objective_diploid_unambig():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta)
obj = poisson.objective(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths)
assert obj < -1e4
def test_poisson_objective_diploid_partially_ambig_biased():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha, beta = -3., 1.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = counts[:, :n] + counts[:, n:]
np.fill_diagonal(counts[:n, :], 0)
np.fill_diagonal(counts[n:, :], 0)
bias = 0.1 + random_state.rand(n)
counts *= np.tile(bias, 2).reshape(-1, 1)
counts *= bias.reshape(-1, 1).T
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta)
obj = poisson.objective(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths,
bias=bias)
assert obj < -1e4
def test_estimate_alpha_beta_diploid_partially_ambig_biased():
lengths = np.array([20])
ploidy = 2
seed = 42
alpha_true, beta_true = -3., 4.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = random_state.rand(n * ploidy, 3)
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta_true * dis**alpha_true
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = counts[:, :n] + counts[:, n:]
np.fill_diagonal(counts[:n, :], 0)
np.fill_diagonal(counts[n:, :], 0)
bias = 0.1 + random_state.rand(n)
counts *= np.tile(bias, 2).reshape(-1, 1)
counts *= bias.reshape(-1, 1).T
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta_true)
alpha, obj, converged, _ = estimate_alpha_beta.estimate_alpha(
X=X_true,
counts=counts,
alpha_init=alpha_true,
lengths=lengths,
bias=bias)
beta = list(
estimate_alpha_beta._estimate_beta(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths,
bias=bias,
verbose=False).values())[0]
assert converged
assert obj < -1e4
assert_array_almost_equal(alpha_true, alpha, decimal=5)
assert_array_almost_equal(beta_true, beta, decimal=3)
def test_mhs_constraint():
lengths = np.array([30])
ploidy = 2
seed = 42
true_interhomo_dis = np.array([10.])
alpha, beta = -3., 1.
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
X_true = np.zeros((n * ploidy, 3), dtype=float)
for i in range(X_true.shape[0]):
X_true[i:, random_state.choice([0, 1, 2])] += 1
X_true[n:] -= X_true[n:].mean(axis=0)
X_true[:n] -= X_true[:n].mean(axis=0)
begin = end = 0
for i in range(len(lengths)):
end += lengths[i]
X_true[begin:end, 0] += true_interhomo_dis[i]
begin = end
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta * dis**alpha
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta)
mhs_k = constraints._mean_interhomolog_counts(counts, lengths=lengths)
constraint = constraints.Constraints(counts,
lengths=lengths,
ploidy=ploidy,
multiscale_factor=1,
constraint_lambdas={'mhs': 1},
constraint_params={'mhs': mhs_k})
constraint.check()
obj = constraint.apply(X_true, alpha=alpha)['obj_mhs']
assert obj < 1e-3
def test_estimate_alpha_beta_diploid_mhs_constraint():
lengths = np.array([30])
ploidy = 2
seed = 42
true_interhomo_dis = np.array([5.])
alpha_true, beta_true = -3., 4.
mhs_lambda = 1
random_state = np.random.RandomState(seed=seed)
n = lengths.sum()
'''X_true = np.zeros((n * ploidy, 3), dtype=float)
for i in range(X_true.shape[0]):
X_true[i:, random_state.choice([0, 1, 2])] += 1'''
X_true = random_state.rand(n * ploidy, 3)
X_true[n:] -= X_true[n:].mean(axis=0)
X_true[:n] -= X_true[:n].mean(axis=0)
begin = end = 0
for i in range(len(lengths)):
end += lengths[i]
X_true[begin:end, 0] += true_interhomo_dis[i]
begin = end
dis = euclidean_distances(X_true)
dis[dis == 0] = np.inf
counts = beta_true * dis**alpha_true
counts[np.isnan(counts) | np.isinf(counts)] = 0
counts = np.triu(counts, 1)
counts = sparse.coo_matrix(counts)
counts = _format_counts(counts=counts,
lengths=lengths,
ploidy=ploidy,
beta=beta_true)
null_counts = [
NullCountsMatrix(counts=counts,
lengths=lengths,
ploidy=ploidy,
multiscale_factor=1)
]
mhs_k = _mean_interhomolog_counts(counts, lengths=lengths)
constraint = Constraints(counts,
lengths=lengths,
ploidy=ploidy,
multiscale_factor=1,
constraint_lambdas={'mhs': mhs_lambda},
constraint_params={'mhs': mhs_k})
constraint.check()
alpha, obj, converged, _ = estimate_alpha_beta.estimate_alpha(
X=X_true,
counts=null_counts,
alpha_init=alpha_true,
lengths=lengths,
constraints=constraint)
beta = list(
estimate_alpha_beta._estimate_beta(X=X_true,
counts=counts,
alpha=alpha,
lengths=lengths,
verbose=False).values())[0]
print(alpha, obj, converged, beta)
assert converged
assert obj < 1e-6
assert_allclose(alpha_true, alpha, rtol=1e-2)
assert_allclose(beta_true, beta, rtol=1e-2)
