def test_gather_residues(self):
methylation = SiteType(name='methylation')
p = Protein(refseq='NM_007', id=1, sequence='ABCD')
sites = [
Site(position=2, types={methylation}, protein=p), # default -> 'B'
Site(position=4, types={methylation}, residue='D'),
]
db.session.add_all(sites)
db.session.commit()
assert methylation.find_modified_residues() == {'B', 'D'}
def train_model(site_type: SiteType,
sequences_dir='.tmp',
sampling_n=10000,
enzyme_type='kinase',
output_path=None,
**kwargs):
"""Train MIMP model for given site type.
NOTE: Natively MIMP works on phosphorylation sites only,
so a special, forked version [reimandlab/rmimp] is needed
for this function to work at all.
Args:
site_type: Type of the site for which the model is to be trained
sequences_dir: path to dir where sequences for trainModel should be dumped
sampling_n: number of sampling iterations for negative sequence set
output_path: path to .mimp file where the model should be saved
**kwargs: will be passed to trainModel
Returns:
trained MIMP model for all kinases affecting sites of given SiteType
"""
if not output_path:
output_path = f'{site_type.name}.mimp'
mimp = load_mimp()
sites_of_this_type = set(site_type.sites)
modified_residues = site_type.find_modified_residues()
negative_sites = gather_negative_sites(modified_residues,
exclude=sites_of_this_type)
sequences_path = Path(sequences_dir)
positive_path = sequences_path / 'positive'
negative_path = sequences_path / 'negative'
for path in [positive_path, negative_path]:
shutil.rmtree(str(path), ignore_errors=True)
path.mkdir(parents=True)
if enzyme_type == 'kinase':
enzymes = Kinase.query.filter(
Kinase.is_involved_in.any(SiteType.name == site_type.name)).filter(
Kinase.sites.any(Site.types.contains(site_type)))
enzymes = tqdm(enzymes, total=enzymes.count())
elif enzyme_type == 'catch-all':
enzymes = [
SimpleNamespace(sites=Site.query.filter(
Site.types.contains(site_type)),
name=f'all_enzymes_for_{site_type.name}')
]
else:
assert False
for enzyme in enzymes:
sites = [site for site in enzyme.sites if site_type in site.types]
positive_sequences = [site.sequence for site in sites]
negative_sequences = sample_random_negative_sequences(
negative_sites, sampling_n)
save_kinase_sequences(enzyme, positive_sequences, positive_path)
save_kinase_sequences(enzyme, negative_sequences, negative_path)
priors = mimp.PRIORS.rx2('human')
# just in case
# r.debug(mimp.trainModel)
return mimp.trainModel(
str(positive_path),
str(negative_path),
file=output_path,
priors=priors, # or calculate_background_frequency(),
# both give the same values (within rounding error), the custom
# func might come in handy in future
residues_groups=residues_groups(site_type, modified_residues),
**kwargs)