def test_scores():
# Create zero row vectors except for 1st and 11th
# (make them have non-zero variance)
x = np.zeros((10, 10), dtype="uint8")
# Make 3rd and 4th perfectly correlated
x[2, :-1] = 1
x[3, :-1] = 1
# Make 8th and 9th partially correlated with 3/4
x[7, :-5] = 1
x[8, :-5] = 1
ds = simulate_genotype_call_dataset(n_variant=x.shape[0],
n_sample=x.shape[1])
ds["dosage"] = (["variants", "samples"], x)
ds = window_by_variant(ds, size=10)
ldm = ld_matrix(ds, threshold=0.2)
idx_drop_ds = maximal_independent_set(ldm)
idx_drop = np.sort(idx_drop_ds.ld_prune_index_to_drop.data)
npt.assert_equal(idx_drop, [3, 8])
# check ld_prune removes correct variants
pruned_ds = ld_prune(ds, threshold=0.2)
npt.assert_equal(pruned_ds.variant_position.values,
[0, 1, 2, 4, 5, 6, 7, 9])
# break tie between 3rd and 4th so 4th wins
scores = np.ones(10, dtype="float32")
scores[2] = 0
scores[3] = 2
ds[variables.variant_score] = (["variants"], scores)
ldm = ld_matrix(ds, threshold=0.2, variant_score=variables.variant_score)
idx_drop_ds = maximal_independent_set(ldm)
idx_drop = np.sort(idx_drop_ds.ld_prune_index_to_drop.data)
npt.assert_equal(idx_drop, [2, 8])
# check ld_prune removes correct variants
pruned_ds = ld_prune(ds,
threshold=0.2,
variant_score=variables.variant_score)
npt.assert_equal(pruned_ds.variant_position.values,
[0, 1, 3, 4, 5, 6, 7, 9])
def test_vs_skallel(args):
x, size, step, threshold, chunks = args
ds = simulate_genotype_call_dataset(n_variant=x.shape[0],
n_sample=x.shape[1])
ds["dosage"] = (["variants", "samples"], da.asarray(x).rechunk({0:
chunks}))
ds = window_by_variant(ds, size=size, step=step)
ldm = ld_matrix(ds, threshold=threshold)
has_duplicates = ldm.compute().duplicated(subset=["i", "j"]).any()
assert not has_duplicates
idx_drop_ds = maximal_independent_set(ldm)
idx_drop = np.sort(idx_drop_ds.ld_prune_index_to_drop.data)
m = allel.locate_unlinked(x, size=size, step=step, threshold=threshold)
idx_drop_ska = np.sort(np.argwhere(~m).squeeze(axis=1))
npt.assert_equal(idx_drop_ska, idx_drop)