def cwm_scan_seqlets(modisco_dir,
output_file,
trim_frac=0.08,
num_workers=1,
contribsf=None,
verbose=False):
"""Compute the cwm scanning scores of the original modisco seqlets
"""
from bpnet.modisco.table import ModiscoData
os.makedirs(os.path.dirname(output_file), exist_ok=True)
add_file_logging(os.path.dirname(output_file), logger, 'cwm_scan_seqlets')
# figure out contrib_wildcard
mf = ModiscoFile(modisco_dir / "modisco.h5")
if contribsf is None:
contrib = ContribFile.from_modisco_dir(modisco_dir)
else:
contrib = contribsf
tasks = mf.tasks()
# HACK prune the tasks of contribution (in case it's present)
tasks = [t.split("/")[0] for t in tasks]
dfi_list = []
for pattern_name in tqdm(mf.pattern_names()):
pattern = mf.get_pattern(pattern_name).trim_seq_ic(trim_frac)
seqlets = mf._get_seqlets(pattern_name, trim_frac=trim_frac)
# scan only the existing locations of the seqlets instead of the full sequences
# to obtain the distribution
stacked_seqlets = contrib.extract(seqlets)
match, contribution = pattern.scan_contribution(
stacked_seqlets.contrib,
hyp_contrib=None,
tasks=tasks,
n_jobs=num_workers,
verbose=False,
pad_mode=None)
seq_match = pattern.scan_seq(stacked_seqlets.seq,
n_jobs=num_workers,
verbose=False,
pad_mode=None)
dfm = pattern.get_instances(tasks,
match,
contribution,
seq_match,
fdr=1,
verbose=verbose,
plot=verbose)
dfm = dfm[dfm.seq_match > 0]
dfi_list.append(dfm)
df = pd.concat(dfi_list)
df.to_csv(output_file)
def modisco_export_patterns(modisco_dir, output_file, contribsf=None):
"""Export patterns to a pkl file. Don't cluster them
Adds `stacked_seqlet_contrib` and `n_seqlets` to pattern `attrs`
Args:
modisco_dir: modisco directory containing
output_file: output file path for patterns.pkl
"""
from bpnet.cli.contrib import ContribFile
logger.info("Loading patterns")
modisco_dir = Path(modisco_dir)
mf = ModiscoFile(modisco_dir / 'modisco.h5')
patterns = [mf.get_pattern(pname) for pname in mf.pattern_names()]
if contribsf is None:
contrib_file = ContribFile.from_modisco_dir(modisco_dir)
logger.info("Loading ContribFile into memory")
contrib_file.cache()
else:
logger.info("Using the provided ContribFile")
contrib_file = contribsf
logger.info("Extracting profile and contribution scores")
extended_patterns = []
for p in tqdm(patterns):
p = p.copy()
# get seqlets
valid_seqlets = mf._get_seqlets(p.name)
# extract the contribution scores
sti = contrib_file.extract(valid_seqlets, profile_width=None)
sti.dfi = mf.get_seqlet_intervals(p.name, as_df=True)
p.attrs['stacked_seqlet_contrib'] = sti
p.attrs['n_seqlets'] = mf.n_seqlets(p.name)
extended_patterns.append(p)
write_pkl(extended_patterns, output_file)