def test_change_quantification():
"""Test if change detection pipeline change at input positions"""
# Run loop change detection between matrices with and without loops
cools = COOLS + COOLS_COMP
conds = ["A"] * len(COOLS) + ["B"] * len(COOLS_COMP)
obs_pos = pah.change_detection_pipeline(cools,
conds,
bed2d_file=str(DATA /
'A_loops.bed2d'),
subsample=False,
percentile_thresh=95)
diff = obs_pos.diff_score
# Check if change was detected in at least half the positions
assert len(diff[diff < 0]) >= len(diff) * 0.5
def test_change_quantification():
"""Test if change detection pipeline change at input positions"""
# Run loop change detection between matrices with and without loops
cools = COOLS + COOLS_COMP
conds = ["B"] * len(COOLS) + ["S"] * len(COOLS_COMP)
obs_pos = pah.change_detection_pipeline(
cools,
conds,
bed2d_file=str(DATA / "B_loops.bed2d"),
subsample=False,
)
diff = obs_pos.diff_score
# Check if change was detected in the correct direction (disappearing)
# some positions
assert len(diff[diff < 0]) >= len(diff) * 0.3
def test_change_no_threshold(kernel):
"""Test if change detection pipeline without threshold reports all input positions"""
# Run loop change detection between matrices with and without loops
cools = COOLS + COOLS_COMP
conds = ["B"] * len(COOLS) + ["S"] * len(COOLS_COMP)
bed2d = pd.read_csv(str(DATA / f"B_{kernel}.bed2d"), sep="\t")
obs_pos = pah.change_detection_pipeline(
cools,
conds,
bed2d_file=str(DATA / f"B_{kernel}.bed2d"),
subsample=False,
pearson_thresh=None,
kernel=kernel,
)
diff = obs_pos.diff_score
# Check if diff scores are returned for all positions
assert len(diff[~np.isnan(diff)]) == bed2d.shape[0]
def test_change_detection():
"""Test if change detection pipeline finds some relevant positions"""
# Run loop change detection between matrices with and without loops
cools = COOLS + COOLS_COMP
conds = ["A"] * len(COOLS) + ["B"] * len(COOLS_COMP)
obs_pos = pah.change_detection_pipeline(cools,
conds,
subsample=False,
percentile_thresh=95)
# Build a set of fuzzy (+/3 pixels around) positions found
valid_pos = set()
for pos in obs_pos.loc[:, ['bin1', 'bin2']].values:
for shift in it.combinations(range(-3, 4), 2):
valid_pos.add((pos[0] + shift[0], pos[1] + shift[1]))
# Count the number of real loop positions that were found
found = 0
for target in LOOPS:
if tuple(target.astype(int)) in valid_pos:
found += 1
assert found / LOOPS.shape[0] >= 0.5
def test_change_detection(kernel, coords):
"""Test if change detection pipeline finds some relevant positions"""
# Run loop change detection between matrices with and without loops
cools = COOLS + COOLS_COMP
conds = ["B"] * len(COOLS) + ["S"] * len(COOLS_COMP)
obs_pos = pah.change_detection_pipeline(cools,
conds,
subsample=False,
kernel=kernel)
# Build a set of fuzzy (+/3 pixels around) positions found
valid_pos = set()
for pos in obs_pos.loc[:, ["bin1", "bin2"]].values:
for shift in it.product(range(-3, 4), range(-3, 4)):
valid_pos.add((pos[0] + shift[0], pos[1] + shift[1]))
# Count the number of real loop positions that were found
found = 0
for target in coords:
if tuple(target.astype(int)) in valid_pos:
found += 1
print(f"Found {found} out of {coords.shape[0]} loops.")
assert found / coords.shape[0] >= 0.5
import cooler
import matplotlib.pyplot as plt
DATA = pathlib.Path("data_test")
# Synthetic matrices and their known loop coordinates
COOLS = [str(c) for c in DATA.glob("B_[1-6]*.cool")]
LOOPS = np.loadtxt(DATA / "B_loops.txt")
# Matrices with a diagonal gradient
COOLS_COMP = [str(c) for c in DATA.glob("smooth_[1-6]*.cool")]
# Run loop change detection between matrices with and without loops
cools = COOLS + COOLS_COMP
conds = ["B"] * len(COOLS) + ["S"] * len(COOLS_COMP)
obs_pos = pah.change_detection_pipeline(
cools,
conds,
min_dist=50000,
subsample=False,
)
# Build a set of fuzzy (+/3 pixels around) positions found
fuzzy_obs = set()
for pos in obs_pos.loc[:, ["bin1", "bin2"]].values:
for shift in it.product(range(-3, 4), range(-3, 4)):
fuzzy_obs.add((pos[0] + shift[0], pos[1] + shift[1]))
# Same for targets
valid_pos = set()
for pos in LOOPS:
for shift in it.product(range(-3, 4), range(-3, 4)):
valid_pos.add((pos[0] + shift[0], pos[1] + shift[1]))
# Count the number of real loop positions that were found
found = 0
for target in LOOPS: