def _get_structure_from_params(params):
predictor = params.get('predictor', "colabfold")
sequence = params.get('sequence')
if not sequence or len(sequence) > 500 or not check_valid_sequence(sequence):
return abort(400, "Sequence is too long or contains invalid characters.")
if predictor not in {"colabfold"}:
return abort(400, "Invalid predictor specified")
ret = get_structure(predictor, sequence)
return ret
def post(self):
params = request.json
sequence = params.get('sequence')
if not sequence or len(sequence) > 2000 or not check_valid_sequence(
sequence):
return abort(
400, "Sequence is too long or contains invalid characters.")
return get_residue_landscape(model_name='prottrans_t5_xl_u50',
sequence=sequence)
def post(self):
params = request.json
sequence = params.get('sequence')
if not sequence or len(sequence) > 2000 or not check_valid_sequence(
sequence):
return abort(
400, "Sequence is too long or contains invalid characters.")
model = {
'seqvec': get_seqvec_embeddings_sync,
'prottrans_bert_bfd': get_protbert_embeddings_sync
}.get(params.get('model', 'seqvec'))
if not model:
return abort(400,
f"Model '{params.get('model')}' isn't available.")
# time_limit && soft_time_limit limit the execution time. Expires limits the queuing time.
job = model.apply_async(args=[sequence],
time_limit=60 * 5,
soft_time_limit=60 * 5,
expires=60 * 60)
embedding = np.array(job.get())
buffer = io.BytesIO()
with h5py.File(buffer, "w") as embeddings_file:
embeddings_file.create_dataset("sequence", data=embedding)
# This simulates closing the file and re-opening it.
# Otherwise the cursor will already be at the end of the
# file when flask tries to read the contents, and it will
# think the file is empty.
buffer.seek(0)
return send_file(buffer,
attachment_filename="embeddings_file.h5",
as_attachment=True)
def post(self):
params = request.json
sequence = params.get('sequence')
if not sequence or len(sequence) > 2000 or not check_valid_sequence(sequence):
return abort(400, "Sequence is too long or contains invalid characters.")
model_name = params.get('model', 'seqvec')
embedding = get_embedding(model_name, sequence)
buffer = io.BytesIO()
with h5py.File(buffer, "w") as embeddings_file:
embeddings_file.create_dataset("sequence", data=embedding)
# This simulates closing the file and re-opening it.
# Otherwise the cursor will already be at the end of the
# file when flask tries to read the contents, and it will
# think the file is empty.
buffer.seek(0)
return send_file(buffer, attachment_filename="embeddings_file.h5", as_attachment=True)
def _get_annotations_from_params(params):
sequence = params.get('sequence')
if not sequence or len(sequence) > 2000 or not check_valid_sequence(
sequence):
return abort(400,
"Sequence is too long or contains invalid characters.")
model_name = params.get('model', 'seqvec')
annotations = get_features(model_name, sequence)
annotations['sequence'] = sequence
format = params.get('format', 'legacy')
only_closest_k = params.get('only_closest_k', True)
if only_closest_k == True:
_filter_ontology(annotations, "predictedBPO")
_filter_ontology(annotations, "predictedCCO")
_filter_ontology(annotations, "predictedMFO")
if format == "protvista-predictprotein":
source = Source(url=request.url,
id="sync",
name=f"bio_embeddings using {model_name}")
evidence = Evidence(source=source, )
protvista_features = dict()
protvista_features['sequence'] = sequence
protvista_features['features'] = list()
if annotations.get('predictedDSSP8'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDSSP8'],
evidences=[evidence],
type=f"SECONDARY_STRUCTURE_8_STATES_({model_name})",
feature_enum=SecondaryStructure))
if annotations.get('predictedDSSP3'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDSSP3'],
evidences=[evidence],
type=f"SECONDARY_STRUCTURE_3_STATES_({model_name})",
feature_enum=SecondaryStructure))
if annotations.get('predictedDSSP3'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDisorder'],
evidences=[evidence],
type=f"DISORDER_({model_name})",
feature_enum=Disorder))
return protvista_features
elif format == "legacy":
predictedCCO = {}
predictedBPO = {}
predictedMFO = {}
for prediction in annotations['predictedCCO']:
predictedCCO[prediction['GO_Term']] = max(
predictedCCO.get(prediction['GO_Term'], -1), prediction['RI'])
for prediction in annotations['predictedBPO']:
predictedBPO[prediction['GO_Term']] = max(
predictedBPO.get(prediction['GO_Term'], -1), prediction['RI'])
for prediction in annotations['predictedMFO']:
predictedMFO[prediction['GO_Term']] = max(
predictedMFO.get(prediction['GO_Term'], -1), prediction['RI'])
annotations['predictedCCO'] = predictedCCO
annotations['predictedBPO'] = predictedBPO
annotations['predictedMFO'] = predictedMFO
return annotations
elif format == "go-predictprotein":
mapping_function = lambda x: {
"gotermid": x['GO_Term'],
"gotermname": x['GO_Name'],
"gotermscore": round(x['RI'] * 100)
}
predictedCCO = {
"ontology":
"Cellular Component Ontology",
"goTermWithScore":
list(map(mapping_function, annotations['predictedCCO']))
}
predictedBPO = {
"ontology":
"Biological Process Ontology",
"goTermWithScore":
list(map(mapping_function, annotations['predictedBPO']))
}
predictedMFO = {
"ontology":
"Molecular Function Ontology",
"goTermWithScore":
list(map(mapping_function, annotations['predictedMFO']))
}
return [predictedBPO, predictedCCO, predictedMFO]
elif format == "full":
return annotations
else:
abort(400, f"Wrong format passed: {format}")
def _get_annotations_from_params(params):
sequence = params.get('sequence')
if not sequence or len(sequence) > 2000 or not check_valid_sequence(
sequence):
return abort(400,
"Sequence is too long or contains invalid characters.")
model_name = params.get('model', 'prottrans_t5_xl_u50')
annotations = get_features(model_name, sequence)
if model_name == 'prottrans_t5_xl_u50' and 'prott5_residue_landscape_annotations' in get_queues(
):
residue_landscape_output = get_residue_landscape(model_name=model_name,
sequence=sequence)
# merge the output of the residue landscape into the feature dict
# add the meta information
for key in residue_landscape_output['meta']:
annotations['meta'][key] = residue_landscape_output['meta'][key]
residue_landscape_output.pop('meta', None)
# add all the remaining information
for key in residue_landscape_output:
annotations[key] = residue_landscape_output[key]
annotations['sequence'] = sequence
format = params.get('format', 'legacy')
only_closest_k = params.get('only_closest_k', True)
if only_closest_k == True:
_filter_ontology(annotations, "predictedBPO")
_filter_ontology(annotations, "predictedCCO")
_filter_ontology(annotations, "predictedMFO")
if format == "protvista-predictprotein":
source = Source(url=request.url,
id="sync",
name=f"bio_embeddings using {model_name}")
evidence = Evidence(source=source, )
protvista_features = dict()
protvista_features['sequence'] = sequence
protvista_features['features'] = list()
if annotations.get('predictedDSSP8'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDSSP8'],
evidences=[evidence],
type=f"SECONDARY_STRUCTURE_8_STATES_({model_name})",
feature_enum=SecondaryStructure))
if annotations.get('predictedDSSP3'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDSSP3'],
evidences=[evidence],
type=f"SECONDARY_STRUCTURE_3_STATES_({model_name})",
feature_enum=SecondaryStructure))
if annotations.get('predictedDisorder'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDisorder'],
evidences=[evidence],
type=f"DISORDER_({model_name})",
feature_enum=Disorder))
if annotations.get('predictedBindingMetal'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedBindingMetal'],
evidences=[evidence],
type=f"BINDING_METAL_({model_name})",
feature_enum=BindingResidues))
if annotations.get('predictedBindingNucleicAcids'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations[
'predictedBindingNucleicAcids'],
evidences=[evidence],
type=f"BINDING_NUCLEIC_ACIDS_({model_name})",
feature_enum=BindingResidues))
if annotations.get('predictedBindingSmallMolecules'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations[
'predictedBindingSmallMolecules'],
evidences=[evidence],
type=f"BINDING_SMALL_MOLECULES_({model_name})",
feature_enum=BindingResidues))
if annotations.get('predictedTransmembrane'):
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedTransmembrane'],
evidences=[evidence],
type=f"TRANSMEMBRANE_({model_name})",
feature_enum=MembraneResidues))
return protvista_features
elif format == "legacy":
predictedCCO = {}
predictedBPO = {}
predictedMFO = {}
for prediction in annotations['predictedCCO']:
predictedCCO[prediction['GO_Term']] = max(
predictedCCO.get(prediction['GO_Term'], -1), prediction['RI'])
for prediction in annotations['predictedBPO']:
predictedBPO[prediction['GO_Term']] = max(
predictedBPO.get(prediction['GO_Term'], -1), prediction['RI'])
for prediction in annotations['predictedMFO']:
predictedMFO[prediction['GO_Term']] = max(
predictedMFO.get(prediction['GO_Term'], -1), prediction['RI'])
annotations['predictedCCO'] = predictedCCO
annotations['predictedBPO'] = predictedBPO
annotations['predictedMFO'] = predictedMFO
return annotations
elif format == "go-predictprotein":
mapping_function = lambda x: {
"gotermid": x['GO_Term'],
"gotermname": x['GO_Name'],
"gotermscore": round(x['RI'] * 100)
}
predictedCCO = {
"ontology":
"Cellular Component Ontology",
"goTermWithScore":
list(map(mapping_function, annotations['predictedCCO']))
}
predictedBPO = {
"ontology":
"Biological Process Ontology",
"goTermWithScore":
list(map(mapping_function, annotations['predictedBPO']))
}
predictedMFO = {
"ontology":
"Molecular Function Ontology",
"goTermWithScore":
list(map(mapping_function, annotations['predictedMFO']))
}
return [predictedBPO, predictedCCO, predictedMFO]
elif format == "full":
return annotations
else:
abort(400, f"Wrong format passed: {format}")
def post(self):
params = request.json
sequence = params.get('sequence')
if not sequence or len(sequence) > 2000 or not check_valid_sequence(
sequence):
return abort(
400, "Sequence is too long or contains invalid characters.")
model_name = params.get('model', 'seqvec')
model = {
'seqvec': get_seqvec_annotations_sync,
'prottrans_bert_bfd': get_protbert_annotations_sync
}.get(model_name)
if not model:
return abort(400, f"Model '{model_name}' isn't available.")
# time_limit && soft_time_limit limit the execution time. Expires limits the queuing time.
job = model.apply_async(args=[sequence],
time_limit=60 * 5,
soft_time_limit=60 * 5,
expires=60 * 60)
annotations = job.get()
annotations['sequence'] = sequence
format = params.get('format', 'legacy')
if format == "protvista-predictprotein":
source = Source(url=request.url,
id="sync",
name=f"bio_embeddings using {model_name}")
evidence = Evidence(source=source, )
protvista_features = dict()
protvista_features['sequence'] = sequence
protvista_features['features'] = list()
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDSSP8'],
evidences=[evidence],
type=f"SECONDARY_STRUCTURE_8_STATES_({model_name})",
feature_enum=SecondaryStructure))
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDSSP3'],
evidences=[evidence],
type=f"SECONDARY_STRUCTURE_3_STATES_({model_name})",
feature_enum=SecondaryStructure))
protvista_features['features'].extend(
annotations_to_protvista_converter(
features_string=annotations['predictedDisorder'],
evidences=[evidence],
type=f"DISORDER_({model_name})",
feature_enum=Disorder))
return protvista_features
else:
return annotations