def _setup_handler(self):
# Only do the rest of this if a query is passed
if self.init_query is not None:
# Setup queries
self._setup_queries()
# Instiatate Reasoners
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
class ChpBrainApiConfig(AppConfig):
logger.warning('Running CHP Brain API Configuration. May take a minute.')
name = 'chp_core_brain'
# Used for distrbuted reasoning
# Get Hosts File if it exists
#parent_dir = os.path.dirname(os.path.realpath(__file__))
#HOSTS_FILENAME = os.path.join(parent_dir, 'hosts')
#NUM_PROCESSES_PER_HOST = multiprocessing.cpu_count()
#if not os.path.exists(HOSTS_FILENAME):
hosts_filename = None
num_processes_per_host = 0
# Instantiate BKB handler
bkb_handler = BkbDataHandler(disease='tcga_gbm',
bkb_major_version='darwin',
bkb_minor_version='2.0')
logger.info('Instantiating reasoners.')
# Instantiate Reasoners
dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=bkb_handler,
hosts_filename=hosts_filename,
num_processes_per_host=num_processes_per_host)
joint_reasoner = ChpJointReasoner(
bkb_handler=bkb_handler,
hosts_filename=hosts_filename,
num_processes_per_host=num_processes_per_host)
def _setup_handler(self):
self.default_survival_target = {
"EFO:0000714": {
"op": '>=',
"value": 970
}
}
# Only do the rest of this if a query is passed
if self.init_query is not None:
# Setup queries
self._setup_queries()
# Instiatate Reasoners
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_handler(self):
# Only do the rest of this if a message is passed
if self.messages is not None:
# Setup messages
self._setup_messages()
# Instiatate Reasoners
if 'default' in self.message_dict:
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
if 'simple' in self.message_dict:
if self.joint_reasoner is None:
self.joint_reasoner = ChpJointReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def setUpClass(cls):
super(TestOneHopHandler, cls).setUpClass()
# load in sample query graphs
with open('query_samples/onehop/standard_queries.json', 'r') as f_:
cls.standard_queries = json.load(f_)
with open('query_samples/onehop/wildcard_queries.json', 'r') as f_:
cls.wildcard_queries = json.load(f_)
cls.bkb_handler = BkbDataHandler()
cls.dynamic_reasoner = ChpDynamicReasoner(cls.bkb_handler)
cls.joint_reasoner = ChpJointReasoner(cls.bkb_handler)
class WildCardHandlerMixin:
def _setup_handler(self):
# Only do the rest of this if a query is passed
if self.init_query is not None:
# Setup queries
self._setup_queries()
# Instiatate Reasoners
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_queries(self):
if type(self.init_query) == list:
self.query_dict = defaultdict(list)
self.query_map = []
for query in self.init_query:
self.query_map.append(query["query_id"])
self.query_dict[self._get_wildcard_type(query)].append(self._setup_single_query(query))
else:
self.query_dict[self._get_wildcard_type(query)].append(self._setup_single_query(query))
def _get_wildcard_type(self, query):
wildcard_type = None
for node_id, node in query["query_graph"]["nodes"].items():
if 'id' not in node:
if wildcard_type is None:
wildcard_type = node['category']
else:
sys.exit('You can only have one contribution target. Make sure to leave only one node with a black curie.')
if wildcard_type == BIOLINK_DRUG:
return 'drug'
elif wildcard_type == BIOLINK_GENE:
return 'gene'
else:
raise ValueError('Did not understand wildcard type {}.'.format(wildcard_type))
def _extract_chp_query(self, query, query_type):
evidence = {}
targets = []
dynamic_evidence = {}
dynamic_targets = {}
# ensure we are using all nodes/edges
total_nodes = 0
total_edges = 0
# get phenotype node
targets = list()
acceptable_target_curies = ['EFO:0000714']
self.implicit_survival_node = False
for node_key in query["query_graph"]['nodes'].keys():
node = query["query_graph"]['nodes'][node_key]
if node['category'] == BIOLINK_PHENOTYPIC_FEATURE and node['id'] in acceptable_target_curies:
target_id = node_key
total_nodes += 1
if total_nodes == 0:
# Use Default Survival
self.implicit_survival_node = True
total_nodes += 1
#acceptable_target_curies_print = ','.join(acceptable_target_curies)
#sys.exit("Survival Node not found. Node category must be '{}' and id must be in: {}".format(BIOLINK_PHENOTYPIC_FEATURE,
# acceptable_target_curies_print))
elif total_nodes > 1:
sys.exit('Too many target nodes')
# get disease node info and ensure only 1 disease:
acceptable_disease_curies = ['MONDO:0007254']
for node_key in query["query_graph"]['nodes'].keys():
node = query["query_graph"]['nodes'][node_key]
if node['category'] == BIOLINK_DISEASE and node['id'] in acceptable_disease_curies:
disease_id = node_key
for edge_key in query["query_graph"]['edges'].keys():
edge = query["query_graph"]['edges'][edge_key]
if edge['predicate'] == BIOLINK_DISEASE_TO_PHENOTYPIC_FEATURE_PREDICATE and edge['subject'] == disease_id and edge['object'] == target_id:
if 'properties' in edge.keys():
days = edge['properties']['days']
qualifier = edge['properties']['qualifier']
else:
days = 970
qualifier = '>='
total_edges += 1
if total_edges > 1:
sys.exit('Disease has too many outgoing edges')
total_nodes += 1
if self.implicit_survival_node:
days=970
qualifier = '>='
total_edges += 1
if total_nodes == 1:
acceptable_disease_curies_print = ','.join(acceptable_disease_curies)
sys.exit("Disease node not found. Node type must be '{}' and curie must be in: {}".format(BIOLINK_DISEASE,
acceptable_disease_curies_print))
elif total_nodes > 2:
sys.exit('Too many disease nodes')
# set BKB target
dynamic_targets['EFO:0000714'] = {
"op": qualifier,
"value": days,
}
truth_target = ('EFO:0000714', '{} {}'.format(qualifier, days))
# get evidence
for node_key in query["query_graph"]['nodes'].keys():
# genes
node = query["query_graph"]['nodes'][node_key]
if node['category'] == BIOLINK_GENE:
# check for appropriate gene node structure
gene_id = node_key
for edge_key in query["query_graph"]['edges'].keys():
edge = query["query_graph"]['edges'][edge_key]
if edge['predicate'] == BIOLINK_GENE_TO_DISEASE_PREDICATE and edge['subject'] == gene_id and edge['object'] == disease_id:
total_edges += 1
if total_edges == total_nodes - 1:
sys.exit("Gene and disease edge not found. Edge type must be '{}'".format(BIOLINK_GENE_TO_DISEASE_PREDICATE))
elif total_edges > total_nodes:
sys.exit('Gene has too many outgoing edges')
# check for appropriate gene node curie
if query_type != 'gene':
gene_curie = node['id']
if gene_curie in self.curies[BIOLINK_GENE]:
gene = gene_curie
else:
sys.exit('Invalid ENSEMBL Identifier. Must be in form ENSEMBL:<ID>.')
evidence["_" + gene] = 'True'
total_nodes += 1
# drugs
if node['category'] == BIOLINK_DRUG:
# check for appropriate drug node structure
drug_id = node_key
for edge_key in query["query_graph"]['edges'].keys():
edge = query["query_graph"]['edges'][edge_key]
if edge['predicate'] == BIOLINK_CHEMICAL_TO_DISEASE_OR_PHENOTYPIC_FEATURE_PREDICATE and edge['subject'] == drug_id and edge['object'] == disease_id:
total_edges += 1
if total_edges == total_nodes - 1:
sys.exit("Drug and disease edge not found. Edge type must be '{}'".format(BIOLINK_CHEMICAL_TO_DISEASE_OR_PHENOTYPIC_FEATURE_PREDICATE))
elif total_edges > total_nodes:
sys.exit('Drug has too many outgoing edges')
# check for appropriate drug node curie
if query_type != 'drug':
drug_curie = node['id']
if drug_curie in self.curies[BIOLINK_DRUG]:
drug = drug_curie
else:
sys.exit('Invalid CHEMBL Identifier: {}. Must be in form CHEMBL:<ID>'.format(drug_curie))
evidence['_' + drug] = 'True'
total_nodes += 1
# Temporary solution to no evidence linking
if len(evidence.keys()) == 0 and len(dynamic_evidence.keys()) == 0:
self.no_evidence_probability_check = True
else:
self.no_evidence_probability_check = False
# produce BKB query
chp_query = Query(
evidence=evidence,
targets=targets,
dynamic_evidence=dynamic_evidence,
dynamic_targets=dynamic_targets,
type='updating')
# Set some other helpful attributes
chp_query.truth_target = truth_target
chp_query.query_id = query["query_id"] if 'query_id' in query else None
return chp_query
def _run_query(self, chp_query, query_type):
""" Runs build BKB query to calculate probability of survival.
A probability is returned to specificy survival time w.r.t a drug.
Contributions for each gene are calculuated and classified under
their true/false target assignments.
"""
# temporary solution to no evidence linking
if not self.no_evidence_probability_check:
if query_type == 'gene':
chp_query = self.dynamic_reasoner.run_query(chp_query, bkb_type='drug')
elif query_type == 'drug':
chp_query = self.dynamic_reasoner.run_query(chp_query, bkb_type='gene')
chp_res_dict = chp_query.result.process_updates()
chp_res_norm_dict = chp_query.result.process_updates(normalize=True)
#chp_query.result.summary()
chp_res_contributions = chp_query.result.process_inode_contributions()
chp_query.truth_prob = max([0, chp_res_norm_dict[chp_query.truth_target[0]][chp_query.truth_target[1]]])
# Collect all source inodes and process patient hashes
patient_contributions = defaultdict(lambda: defaultdict(int))
for target, contrib_dict in chp_res_contributions.items():
target_comp_name, target_state_name = target
for inode, contrib in contrib_dict.items():
comp_name, state_name = inode
if '_Source_' in comp_name:
# Split source state name to get patient hashes
source_hashes_str = state_name.split('_')[-1]
source_hashes = [int(source_hash) for source_hash in source_hashes_str.split(',')]
hash_len = len(source_hashes)
# Process patient contributions
for _hash in source_hashes:
# Normalize to get relative contribution
patient_contributions[target][_hash] += contrib/hash_len #/ chp_res_dict[target_comp_name][target_state_name]
else:
# probability of survival
num_survived = 0
num_all = len(self.dynamic_reasoner.raw_patient_data.keys())
str_op = chp_query.dynamic_targets['EFO:0000714']['op']
opp_op = get_opposite_operator(str_op)
op = get_operator(str_op)
days = chp_query.dynamic_targets['EFO:0000714']['value']
for patient, pat_dict in self.dynamic_reasoner.raw_patient_data.items():
if op(pat_dict['survival_time'], days):
num_survived += 1
chp_query.truth_prob = num_survived/num_all
# patient_contributions
patient_contributions = defaultdict(lambda: defaultdict(int))
for patient, pat_dict in self.dynamic_reasoner.raw_patient_data.items():
if op(pat_dict['survival_time'], days):
if num_survived == 0:
patient_contributions[('EFO:0000714', '{} {}'.format(str_op, days))][patient] = 0
else:
patient_contributions[('EFO:0000714', '{} {}'.format(str_op, days))][patient] = chp_query.truth_prob/num_survived
else:
if num_survived == 0:
patient_contributions[('EFO:0000714', '{} {}'.format(opp_op, days))][patient] = (1-chp_query.truth_prob)/num_all
else:
patient_contributions[('EFO:0000714', '{} {}'.format(opp_op, days))][patient] = (1-chp_query.truth_prob)/(num_all-num_survived)
# Now iterate through the patient data to translate patient contributions to drug/gene contributions
wildcard_contributions = defaultdict(lambda: defaultdict(int))
for target, patient_contrib_dict in patient_contributions.items():
for patient, contrib in patient_contrib_dict.items():
if query_type == 'gene':
for gene_curie in self.dynamic_reasoner.raw_patient_data[patient]["gene_curies"]:
wildcard_contributions[gene_curie][target] += contrib
elif query_type == 'drug':
for drug_curie in self.dynamic_reasoner.raw_patient_data[patient]["drug_curies"]:
wildcard_contributions[drug_curie][target] += contrib
# normalize gene contributions by the target and take relative difference
for curie in wildcard_contributions.keys():
truth_target_gene_contrib = 0
nontruth_target_gene_contrib = 0
for target, contrib in wildcard_contributions[curie].items():
if target[0] == chp_query.truth_target[0] and target[1] == chp_query.truth_target[1]:
truth_target_gene_contrib += contrib / chp_query.truth_prob
else:
nontruth_target_gene_contrib += contrib / (1 - chp_query.truth_prob)
wildcard_contributions[curie]['relative'] = truth_target_gene_contrib - nontruth_target_gene_contrib
chp_query.report = None
chp_query.wildcard_contributions = wildcard_contributions
return chp_query
def _construct_trapi_response(self, chp_query, query_type):
# Get orginal query
if len(self.init_query) == 1:
query = self.init_query[0]
query_id = None
else:
for _query in self.init_query:
if _query["query_id"] == chp_query.query_id:
query = _query
query_id = query["query_id"]
break
# Construct first result which is the result of the standard probablistic query.
kg = copy.deepcopy(query["query_graph"])
# Process Nodes
node_pairs = defaultdict(None)
contrib_qg_id = None
for node_key in list(kg["nodes"].keys())[:]:
qg_node_curie = kg['nodes'][node_key].pop('id', None)
if qg_node_curie is not None:
kg['nodes'][qg_node_curie] = kg['nodes'].pop(node_key)
if kg['nodes'][qg_node_curie]['category'] == BIOLINK_GENE:
kg['nodes'][qg_node_curie]['name'] = self.curies["biolink:Gene"][qg_node_curie][0]
elif kg['nodes'][qg_node_curie]['category'] == BIOLINK_DRUG:
kg['nodes'][qg_node_curie]['name'] = self.curies["biolink:Drug"][qg_node_curie][0]
node_pairs[node_key] = qg_node_curie
else:
kg["nodes"].pop(node_key)
if not self.implicit_survival_node:
# Process Edges
edge_pairs = dict()
knowledge_edges = 0
for edge_key in list(kg['edges'].keys())[:]:
subject_node = kg['edges'][edge_key]['subject']
if kg['edges'][edge_key]['predicate'] == BIOLINK_GENE_TO_DISEASE_PREDICATE and query['query_graph']['nodes'][subject_node]['category'] == BIOLINK_GENE and query_type == 'gene':
kg['edges'].pop(edge_key)
elif kg['edges'][edge_key]['predicate'] == BIOLINK_CHEMICAL_TO_DISEASE_OR_PHENOTYPIC_FEATURE_PREDICATE and query['query_graph']['nodes'][subject_node]['category'] == BIOLINK_DRUG and query_type == 'drug':
kg['edges'].pop(edge_key)
else:
kg_id = 'kge{}'.format(knowledge_edges)
knowledge_edges += 1
kg['edges'][kg_id] = kg['edges'].pop(edge_key)
kg['edges'][kg_id]['subject'] = node_pairs[kg['edges'][kg_id]['subject']]
kg['edges'][kg_id]['object'] = node_pairs[kg['edges'][kg_id]['object']]
edge_pairs[edge_key] = kg_id
if kg['edges'][kg_id]['predicate'] == BIOLINK_DISEASE_TO_PHENOTYPIC_FEATURE_PREDICATE:
if 'properties' in kg['edges'][kg_id].keys():
kg['edges'][kg_id].pop('properties')
kg['edges'][kg_id]['attributes'] = [{'name':'Probability of Survival',
'type':BIOLINK_PROBABILITY,
'value':chp_query.truth_prob}]
# Put first result of standard prob query of only curie nodes (i.e. no wildcard nodes where used as evidence)
results = []
results.append({'edge_bindings':dict(),
'node_bindings':dict()})
for edge_pair_key in edge_pairs:
results[0]['edge_bindings'][edge_pair_key] = [{ 'id': str(edge_pairs[edge_pair_key])}]
for node_pair_key in node_pairs:
results[0]['node_bindings'][node_pair_key] = [{ 'id': str(node_pairs[node_pair_key])}]
else:
knowledge_edges = 0
kg['edges'] = {}
results = []
# Build relative contribution results and added associated edges into knowledge graph
unsorted_wildcard_contributions = []
for wildcard, contrib_dict in chp_query.wildcard_contributions.items():
unsorted_wildcard_contributions.append((contrib_dict['relative'], wildcard))
sorted_wildcard_contributions = [(contrib,wildcard) for contrib, wildcard in sorted(unsorted_wildcard_contributions, key=lambda x: abs(x[0]), reverse=True)]
for contrib, wildcard in sorted_wildcard_contributions[:self.max_results]:
rg = copy.deepcopy(query["query_graph"])
_node_pairs = {}
_edge_pairs = {}
# Process node pairs
for node_id, node in rg["nodes"].items():
if node["category"] == BIOLINK_GENE and query_type == 'gene':
kg["nodes"][wildcard] = copy.deepcopy(node)
kg["nodes"][wildcard].update({"name": self.curies[BIOLINK_GENE][wildcard][0]})
_node_pairs[node_id] = wildcard
elif node["category"] == BIOLINK_DRUG and query_type == 'drug':
kg["nodes"][wildcard] = copy.deepcopy(node)
kg["nodes"][wildcard].update({"name": self.curies[BIOLINK_DRUG][wildcard][0]})
_node_pairs[node_id] = wildcard
else:
_node_pairs[node_id] = node_pairs[node_id]
# Process edge pairs
for edge_id, edge in rg["edges"].items():
subject_node = edge['subject']
if query_type == 'gene' and edge["predicate"] == BIOLINK_GENE_TO_DISEASE_PREDICATE and query['query_graph']['nodes'][subject_node]['category'] == BIOLINK_GENE:
knowledge_edges += 1
kg_edge_id = 'kge{}'.format(knowledge_edges)
kg["edges"][kg_edge_id] = copy.deepcopy(edge)
kg["edges"][kg_edge_id]["subject"] = _node_pairs[kg["edges"][kg_edge_id]["subject"]]
kg["edges"][kg_edge_id]["object"] = _node_pairs[kg["edges"][kg_edge_id]["object"]]
#kg["edges"][kg_edge_id]["value"] = contrib
kg["edges"][kg_edge_id]["attributes"] = [{'name':'Contribution',
'type':BIOLINK_CONTRIBUTION,
'value':contrib}]
_edge_pairs[edge_id] = kg_edge_id
elif query_type == 'drug' and edge["predicate"] == BIOLINK_CHEMICAL_TO_DISEASE_OR_PHENOTYPIC_FEATURE_PREDICATE and query['query_graph']['nodes'][subject_node]['category'] == BIOLINK_DRUG:
knowledge_edges += 1
kg_edge_id = 'kge{}'.format(knowledge_edges)
kg["edges"][kg_edge_id] = copy.deepcopy(edge)
kg["edges"][kg_edge_id]["subject"] = _node_pairs[kg["edges"][kg_edge_id]["subject"]]
kg["edges"][kg_edge_id]["object"] = _node_pairs[kg["edges"][kg_edge_id]["object"]]
#kg["edges"][kg_edge_id]["value"] = contrib
kg["edges"][kg_edge_id]["attributes"] = [{'name':'Contribution',
'type':BIOLINK_CONTRIBUTION,
'value':contrib}]
_edge_pairs[edge_id] = kg_edge_id
else:
_edge_pairs[edge_id] = edge_pairs[edge_id]
# Process node and edge binding results
_res = {"edge_bindings": {},
"node_bindings": {}}
for edge_pair_key in _edge_pairs:
_res["edge_bindings"][edge_pair_key] = [{ "id": str(_edge_pairs[edge_pair_key])}]
for node_pair_key in _node_pairs:
_res["node_bindings"][node_pair_key] = [{ "id": str(_node_pairs[node_pair_key])}]
results.append(_res)
# query response
trapi_message = {'query_graph': query["query_graph"],
'knowledge_graph': kg,
'results': results}
trapi_response = {'message' : trapi_message}
return query_id, trapi_response
class OneHopHandlerMixin:
""" OneHopeHandler is the handler for 1-hop queries. That is
query graphs (QGs) that consists of 2 nodes and a single edge.
:param query: the query graph sent by the ARA.
:type query: dict
:param hosts_filename: a filename for a stored QG. Defaults to None
:type hosts_filename: str
:param num_processes_per_host: Not implemented thouroughly, but would be
used for distributed reasoning.
:type num_processes_per_host: int
:param max_results: specific to 1-hop queries, specifies the number of
wildcard genes to return.
:type max_results: int
"""
def _setup_handler(self):
self.default_survival_target = {
"EFO:0000714": {
"op": '>=',
"value": 970
}
}
# Only do the rest of this if a query is passed
if self.init_query is not None:
# Setup queries
self._setup_queries()
# Instiatate Reasoners
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_queries(self):
if type(self.init_query) == list:
self.query_dict = defaultdict(list)
self.query_map = []
for query in self.init_query:
self.query_map.append(query["query_id"])
self.query_dict[self._get_wildcard_type(query)].append(
self._setup_single_query(query))
else:
self.query_dict[self._get_wildcard_type(query)].append(
self._setup_single_query(query))
def _get_wildcard_type(self, query):
wildcard_type = None
for node_id, node in query["query_graph"]["nodes"].items():
if 'id' not in node:
if wildcard_type is None:
wildcard_type = node['category']
else:
sys.exit(
'You can only have one contribution target. Make sure to leave only one node with a black curie.'
)
if wildcard_type == BIOLINK_DRUG:
return 'drug'
elif wildcard_type == BIOLINK_GENE:
return 'gene'
else:
raise ValueError(
'Did not understand wildcard type {}.'.format(wildcard_type))
def check_query(self):
""" Currently not implemented. Would check validity of query.
"""
return True
def _extract_chp_query(self, query, query_type=None):
evidence = {}
dynamic_targets = {}
if len(query["query_graph"]['nodes']) > 2 or len(
query["query_graph"]['edges']) > 1:
sys.exit('1 hop quries can only have 2 nodes and 1 edge')
# check edge for source and target
edge_key = list(query["query_graph"]["edges"].keys())[0]
edge = query["query_graph"]['edges'][edge_key]
if 'subject' not in edge.keys() or 'object' not in edge.keys():
sys.exit(
'Edge must have both a \'subject\' and and \'object\' key')
subject = edge['subject']
obj = edge['object']
# Get non-wildcard node
if query_type == 'gene':
if query["query_graph"]['nodes'][subject][
'category'] != BIOLINK_GENE:
sys.exit('Subject node must be \'category\' {}'.format(
BIOLINK_GENE))
drug_curie = query["query_graph"]['nodes'][obj]['id']
if drug_curie not in self.curies[BIOLINK_DRUG]:
sys.exit('Invalid CHEMBL Identifier. Must be CHEMBL:<ID>')
evidence['_{}'.format(drug_curie)] = 'True'
elif query_type == 'drug':
if query["query_graph"]['nodes'][subject][
'category'] != BIOLINK_DRUG:
sys.exit('Subject node must be \'category\' {}'.format(
BIOLINK_DRUG))
gene_curie = query["query_graph"]['nodes'][obj]['id']
if gene_curie not in self.curies[BIOLINK_GENE]:
sys.exit('Invalid ENSEMBL Identifier. Must be ENSEMBL:<ID>')
evidence['_{}'.format(gene_curie)] = 'True'
# default survival time
dynamic_targets.update(self.default_survival_target)
truth_target = ('EFO:0000714', '{} {}'.format(
self.default_survival_target["EFO:0000714"]["op"],
self.default_survival_target["EFO:0000714"]["value"]))
chp_query = Query(evidence=evidence,
targets=None,
dynamic_evidence=None,
dynamic_targets=dynamic_targets,
type='updating')
# Set some other helpful attributes
chp_query.truth_target = truth_target
chp_query.query_id = query["query_id"] if 'query_id' in query else None
return chp_query
def _run_query(self, chp_query, query_type):
""" Runs build BKB query to calculate probability of survival.
A probability is returned to specificy survival time w.r.t a drug.
Contributions for each gene are calculuated and classified under
their true/false target assignments.
"""
if query_type == 'gene':
chp_query = self.dynamic_reasoner.run_query(chp_query,
bkb_type='drug')
elif query_type == 'drug':
chp_query = self.dynamic_reasoner.run_query(chp_query,
bkb_type='gene')
chp_res_dict = chp_query.result.process_updates()
chp_res_norm_dict = chp_query.result.process_updates(normalize=True)
#chp_query.result.summary()
chp_res_contributions = chp_query.result.process_inode_contributions()
chp_query.truth_prob = max([
0, chp_res_norm_dict[chp_query.truth_target[0]][
chp_query.truth_target[1]]
])
#print(chp_res_contributions)
# Collect all source inodes and process patient hashes
patient_contributions = defaultdict(lambda: defaultdict(int))
for target, contrib_dict in chp_res_contributions.items():
target_comp_name, target_state_name = target
for inode, contrib in contrib_dict.items():
comp_name, state_name = inode
if '_Source_' in comp_name:
# Split source state name to get patient hashes
source_hashes_str = state_name.split('_')[-1]
source_hashes = [
int(source_hash)
for source_hash in source_hashes_str.split(',')
]
hash_len = len(source_hashes)
# Process patient contributions
for _hash in source_hashes:
# Normalize to get relative contribution
patient_contributions[target][
_hash] += contrib / hash_len #/ chp_res_dict[target_comp_name][target_state_name]
# Now iterate through the patient data to translate patient contributions to drug/gene contributions
wildcard_contributions = defaultdict(lambda: defaultdict(int))
for target, patient_contrib_dict in patient_contributions.items():
for patient, contrib in patient_contrib_dict.items():
if query_type == 'gene':
for gene_curie in self.dynamic_reasoner.raw_patient_data[
patient]["gene_curies"]:
wildcard_contributions[gene_curie][target] += contrib
elif query_type == 'drug':
for drug_curie in self.dynamic_reasoner.raw_patient_data[
patient]["drug_curies"]:
wildcard_contributions[drug_curie][target] += contrib
# normalize gene contributions by the target and take relative difference
for curie in wildcard_contributions.keys():
truth_target_gene_contrib = 0
nontruth_target_gene_contrib = 0
for target, contrib in wildcard_contributions[curie].items():
if target[0] == chp_query.truth_target[0] and target[
1] == chp_query.truth_target[1]:
truth_target_gene_contrib += contrib / chp_res_dict[
target[0]][target[1]]
else:
nontruth_target_gene_contrib += contrib / chp_res_dict[
target[0]][target[1]]
wildcard_contributions[curie][
'relative'] = truth_target_gene_contrib - nontruth_target_gene_contrib
chp_query.report = None
chp_query.wildcard_contributions = wildcard_contributions
return chp_query
def _construct_trapi_response(self, chp_query, query_type):
# Get orginal query
if len(self.init_query) == 1:
query = self.init_query[0]
query_id = None
else:
for _query in self.init_query:
if _query["query_id"] == chp_query.query_id:
query = _query
query_id = query["query_id"]
break
kg = copy.deepcopy(query["query_graph"])
edge_bindings = {}
node_bindings = {}
# get edge subject, object, edge label and pop edge
edge_key = list(kg['edges'].keys())[0]
edge = kg['edges'][edge_key]
edge_label = edge['predicate']
subject = edge['subject']
obj = edge['object']
kg['edges'].pop(edge_key)
# move curie to key
non_wildcard_curie = kg['nodes'][obj].pop('id')
kg['nodes'][non_wildcard_curie] = kg['nodes'].pop(obj)
if query_type == 'gene':
kg['nodes'][non_wildcard_curie]['name'] = self._get_curie_name(
BIOLINK_DRUG, non_wildcard_curie)[0]
elif query_type == 'drug':
kg['nodes'][non_wildcard_curie]['name'] = self._get_curie_name(
BIOLINK_GENE, non_wildcard_curie)[0]
node_bindings[obj] = non_wildcard_curie
# remove wildcard gene node from kg
kg['nodes'].pop(subject)
# Build relative contribution results and added associated edges into knowledge graph
unsorted_wildcard_contributions = []
for wildcard, contrib_dict in chp_query.wildcard_contributions.items():
unsorted_wildcard_contributions.append(
(contrib_dict['relative'], wildcard))
sorted_wildcard_contributions = [
(contrib, wildcard)
for contrib, wildcard in sorted(unsorted_wildcard_contributions,
key=lambda x: abs(x[0]),
reverse=True)
]
# add kg gene nodes and edges
edge_count = 0
node_count = 1
results = []
for contrib, wildcard in sorted_wildcard_contributions[:self.
max_results]:
if query_type == 'gene':
kg['nodes'][wildcard] = {
"name": self._get_curie_name(BIOLINK_GENE, wildcard)[0],
"category": BIOLINK_GENE
}
# add edge
kg['edges']['kge{}'.format(edge_count)] = {
"predicate":
BIOLINK_CHEMICAL_TO_GENE_PREDICATE,
"subject":
wildcard,
"object":
non_wildcard_curie,
"attributes": [{
'name': 'Contribution',
'type': BIOLINK_CONTRIBUTION,
'value': contrib
}]
}
elif query_type == 'drug':
kg['nodes'][wildcard] = {
"name": self._get_curie_name(BIOLINK_DRUG, wildcard)[0],
"category": BIOLINK_DRUG
}
# add edge
kg['edges']['kge{}'.format(edge_count)] = {
"predicate":
BIOLINK_CHEMICAL_TO_GENE_PREDICATE,
"subject":
wildcard,
"object":
non_wildcard_curie,
"attributes": [{
'name': 'Contribution',
'type': BIOLINK_CONTRIBUTION,
'value': contrib
}]
}
# add to results
node_binding = {
obj: [{
'id': non_wildcard_curie
}],
subject: [{
'id': wildcard
}]
}
edge_binding = {edge_key: [{'id': 'kge{}'.format(edge_count)}]}
results.append({
'node_bindings': node_binding,
'edge_bindings': edge_binding
})
edge_count += 1
node_count += 1
# query response
trapi_message = {
'query_graph': query["query_graph"],
'knowledge_graph': kg,
'results': results
}
trapi_response = {'message': trapi_message}
return query_id, trapi_response
class DefaultHandlerMixin:
def _setup_handler(self):
# Only do the rest of this if a message is passed
if self.messages is not None:
# Setup messages
self._setup_messages()
# Instiatate Reasoners
if 'default' in self.message_dict:
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
if 'simple' in self.message_dict:
if self.joint_reasoner is None:
self.joint_reasoner = ChpJointReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_messages(self):
self.message_dict = defaultdict(list)
for message in self.messages:
if self._is_simple_message(message):
self.message_dict['simple'].append(message)
else:
self.message_dict['default'].append(message)
def _is_simple_message(self, message):
""" Check if this is a {0 or 1} drug, {0 or 1} gene, one outcome standard message.
"""
_found_outcome = False
_found_disease = False
_found_gene = False
_found_drug = False
query_graph = message.query_graph
for node_key, node in query_graph.nodes.items():
if node.categories[0] == BIOLINK_PHENOTYPIC_FEATURE_ENTITY:
# If we've already found the target and there's another phenotypic feature, then this isn't simple.
if _found_outcome:
return False
else:
_found_outcome = True
if node.categories[0] == BIOLINK_DISEASE_ENTITY:
# If we've already found disease and there's another disease, then this isn't simple.
if _found_disease:
return False
else:
_found_disease = True
if node.categories[0] == BIOLINK_GENE_ENTITY:
if _found_gene:
return False
else:
_found_gene = True
if node.categories[0] == BIOLINK_DRUG_ENTITY:
if _found_drug:
return False
else:
_found_drug = True
return True
def _extract_chp_query(self, message, message_type=None):
# Initialize Chp Query
chp_query = ChpQuery(reasoning_type='updating')
# Ensure we are using all nodes/edges
total_nodes = 0
total_edges = 0
query_graph = message.query_graph
# get phenotype node
targets = list()
for node_key in query_graph.nodes.keys():
node = query_graph.nodes[node_key]
if node.categories[0] == BIOLINK_PHENOTYPIC_FEATURE_ENTITY:
target_id = node_key
total_nodes += 1
survival_value = 970
survival_operator = '>='
# get disease node info and ensure only 1 disease:
for node_key in query_graph.nodes.keys():
node = query_graph.nodes[node_key]
if node.categories[0] == BIOLINK_DISEASE_ENTITY:
disease_id = node_key
for edge_key in query_graph.edges.keys():
edge = query_graph.edges[edge_key]
if self.check_predicate_support(
edge.predicates[0], BIOLINK_HAS_PHENOTYPE_ENTITY
) and edge.subject == disease_id and edge.object == target_id:
survival_time_constraint = edge.find_constraint(
name='survival_time')
if survival_time_constraint is not None:
survival_value = survival_time_constraint.value
survival_operator = survival_time_constraint.operator
if survival_operator == 'matches':
survival_operator = '=='
total_edges += 1
total_nodes += 1
# set BKB target
chp_query.add_dynamic_target(node.ids[0], survival_operator,
survival_value)
truth_target = (node.ids[0], '{} {}'.format(survival_operator,
survival_value))
# get evidence
for node_key in query_graph.nodes.keys():
# genes
node = query_graph.nodes[node_key]
if node.categories[0] == BIOLINK_GENE_ENTITY:
# check for appropriate gene node structure
gene_id = node_key
for edge_key in query_graph.edges.keys():
edge = query_graph.edges[edge_key]
if self.check_predicate_support(
edge.predicates[0],
BIOLINK_GENE_ASSOCIATED_WITH_CONDITION_ENTITY
) and edge.subject == gene_id and edge.object == disease_id:
total_edges += 1
# check for appropriate gene node curie
gene_curie = node.ids[0]
gene = gene_curie
chp_query.add_meta_evidence(gene, 'True')
total_nodes += 1
# drugs
if node.categories[0] == BIOLINK_DRUG_ENTITY:
# check for appropriate drug node structure
drug_id = node_key
for edge_key in query_graph.edges.keys():
edge = query_graph.edges[edge_key]
if self.check_predicate_support(
edge.predicates[0], BIOLINK_TREATS_ENTITY
) and edge.subject == drug_id and edge.object == disease_id:
total_edges += 1
# check for appropriate drug node curie
drug_curie = node.ids[0]
drug = drug_curie
chp_query.add_dynamic_evidence(node.ids[0], '==', 'True')
total_nodes += 1
# Set some other helpful attributes
chp_query.truth_target = truth_target
return chp_query
def _run_query(self, chp_query, query_type):
if query_type == 'simple':
chp_query = self.joint_reasoner.run_query(chp_query)
# If a probability was found for the target
if len(chp_query.result) > 0:
# If a probability was found for the truth target
if chp_query.truth_target in chp_query.result:
total_unnormalized_prob = 0
for target, contrib in chp_query.result.items():
prob = max(0, contrib)
total_unnormalized_prob += prob
chp_query.truth_prob = max([
0, chp_query.result[(chp_query.truth_target)]
]) / total_unnormalized_prob
else:
chp_query.truth_prob = 0
else:
chp_query.truth_prob = -1
chp_query.report = None
else:
chp_query = self.dynamic_reasoner.run_query(chp_query)
chp_res_dict = chp_query.result.process_updates(normalize=True)
try:
chp_query.truth_prob = max([
0, chp_res_dict[chp_query.truth_target[0]][
chp_query.truth_target[1]]
])
except KeyError:
# May need to come back and fix this.
chp_query.truth_prob = -1
chp_query.report = None
return chp_query
def _construct_trapi_message(self, chp_query, message, query_type=None):
# update target node info and form edge pair combos for results graph
qg = message.query_graph
kg = message.knowledge_graph
node_bindings = {}
for qnode_key, qnode in qg.nodes.items():
if qnode.categories[0] == BIOLINK_GENE_ENTITY:
knode_key = kg.add_node(
qnode.ids[0],
self.curies[BIOLINK_GENE_ENTITY.get_curie()][qnode.ids[0]]
[0],
qnode.categories[0].get_curie(),
)
elif qnode.categories[0] == BIOLINK_DRUG_ENTITY:
knode_key = kg.add_node(
qnode.ids[0],
self.curies[BIOLINK_DRUG_ENTITY.get_curie()][qnode.ids[0]]
[0],
qnode.categories[0].get_curie(),
)
else:
knode_key = kg.add_node(
qnode.ids[0],
qnode.ids[0],
qnode.categories[0].get_curie(),
)
node_bindings[qnode_key] = [knode_key]
edge_bindings = {}
for qedge_key, qedge in qg.edges.items():
kedge_key = kg.add_edge(
node_bindings[qedge.subject][0],
node_bindings[qedge.object][0],
predicate=qedge.predicates[0].get_curie(),
relation=qedge.relation,
)
edge_bindings[qedge_key] = [kedge_key]
# Add Attribute
if self.check_predicate_support(qedge.predicates[0],
BIOLINK_HAS_PHENOTYPE_ENTITY):
kg.edges[kedge_key].add_attribute(
attribute_type_id='Probability of Survival',
value=chp_query.truth_prob,
value_type_id=BIOLINK_HAS_CONFIDENCE_LEVEL_ENTITY.
get_curie(),
)
# Proces results
message.results.add_result(
node_bindings,
edge_bindings,
)
return message
def setUp(self):
self.bkb_handler = BkbDataHandler(
bkb_major_version='coulomb',
bkb_minor_version='1.0',
)
self.dynamic_reasoner = ChpDynamicReasoner(self.bkb_handler)
class TestDynamicReasoner(unittest.TestCase):
def setUp(self):
self.bkb_handler = BkbDataHandler(
bkb_major_version='coulomb',
bkb_minor_version='1.0',
)
self.dynamic_reasoner = ChpDynamicReasoner(self.bkb_handler)
def test_dynamic_reasoner_one_gene(self):
# Specify evidence
evidence = {'_ENSEMBL:ENSG00000155657': 'True'}
# Specify targets
dynamic_targets = {"EFO:0000714": {"op": '>=', "value": 1000}}
# Setup query
query = Query(evidence=evidence, dynamic_targets=dynamic_targets)
query = self.dynamic_reasoner.run_query(query)
query.result.summary(include_contributions=False)
def test_dynamic_reasoner_one_gene_one_drug(self):
# Specify evidence
evidence = {
'_ENSEMBL:ENSG00000155657': 'True',
'CHEMBL:CHEMBL83': 'True',
}
# Specify targets
dynamic_targets = {"EFO:0000714": {"op": '>=', "value": 1000}}
# Setup query
query = Query(evidence=evidence, dynamic_targets=dynamic_targets)
query = self.dynamic_reasoner.run_query(query)
query.result.summary(include_contributions=False)
def test_dynamic_reasoner_two_gene_one_drug(self):
# Specify evidence
evidence = {
'_ENSEMBL:ENSG00000155657': 'True',
'_ENSEMBL:ENSG00000241973': 'True',
'CHEMBL:CHEMBL83': 'True',
}
# Specify targets
dynamic_targets = {"EFO:0000714": {"op": '>=', "value": 1000}}
# Setup query
query = Query(evidence=evidence, dynamic_targets=dynamic_targets)
query = self.dynamic_reasoner.run_query(query)
query.result.summary(include_contributions=False)
def test_dynamic_reasoner_one_drug_survival(self):
# Specify evidence
evidence = {
'_CHEMBL:CHEMBL83': 'True',
}
# Specify targets
dynamic_targets = {"EFO:0000714": {"op": '>=', "value": 1000}}
# Setup query
query = Query(evidence=evidence, dynamic_targets=dynamic_targets)
query = self.dynamic_reasoner.run_query(query, bkb_type='drug')
query.result.summary(include_contributions=False)
def test_dynamic_reasoner_two_drug_survival(self):
# Specify evidence
evidence = {
'_CHEMBL:CHEMBL83': 'True',
'_CHEMBL:CHEMBL1201247': 'True',
}
# Specify targets
dynamic_targets = {"EFO:0000714": {"op": '>=', "value": 1000}}
# Setup query
query = Query(evidence=evidence, dynamic_targets=dynamic_targets)
query = self.dynamic_reasoner.run_query(query, bkb_type='drug')
query.result.summary(include_contributions=False)
class DefaultHandlerMixin:
def _setup_handler(self):
# Only do the rest of this if a query is passed
if self.init_query is not None:
# Setup queries
self._setup_queries()
# Instiatate Reasoners
if 'default' in self.query_dict:
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
if 'simple' in self.query_dict:
if self.joint_reasoner is None:
self.joint_reasoner = ChpJointReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_queries(self):
if type(self.init_query) == list:
self.query_dict = defaultdict(list)
self.query_map = []
for query in self.init_query:
self.query_map.append(query["query_id"])
if self._is_simple_query(query):
self.query_dict['simple'].append(
self._setup_single_query(query))
else:
self.query_dict['default'].append(
self._setup_single_query(query))
else:
if self._is_simple_query(self.init_query):
self.query_dict = {
"simple": [self._setup_single_query(self.init_query)]
}
else:
self.query_dict = {
"default": [self._setup_single_query(self.init_query)]
}
def _is_simple_query(self, query):
""" Check if this is a {0 or 1} drug, {0 or 1} gene, one outcome standard query.
"""
_found_outcome = False
_found_disease = False
_found_gene = False
_found_drug = False
for node_key, node in query["query_graph"]["nodes"].items():
if node["category"] == BIOLINK_PHENOTYPIC_FEATURE:
# If we've already found the target and there's another phenotypic feature, then this isn't simple.
if _found_outcome:
return False
else:
_found_outcome = True
if node['category'] == BIOLINK_DISEASE:
# If we've already found disease and there's another disease, then this isn't simple.
if _found_disease:
return False
else:
_found_disease = True
if node["category"] == BIOLINK_GENE:
if _found_gene:
return False
else:
_found_gene = True
if node['category'] == BIOLINK_DRUG:
if _found_drug:
return False
else:
_found_drug = True
return True
def _extract_chp_query(self, query, query_type=None):
evidence = {}
targets = []
dynamic_evidence = {}
dynamic_targets = {}
# ensure we are using all nodes/edges
total_nodes = 0
total_edges = 0
# get phenotype node
targets = list()
for node_key in query["query_graph"]['nodes'].keys():
node = query["query_graph"]['nodes'][node_key]
if node['category'] == BIOLINK_PHENOTYPIC_FEATURE:
target_id = node_key
total_nodes += 1
# get disease node info and ensure only 1 disease:
for node_key in query["query_graph"]['nodes'].keys():
node = query["query_graph"]['nodes'][node_key]
if node['category'] == BIOLINK_DISEASE:
disease_id = node_key
for edge_key in query["query_graph"]['edges'].keys():
edge = query["query_graph"]['edges'][edge_key]
if edge['predicate'] == BIOLINK_DISEASE_TO_PHENOTYPIC_FEATURE_PREDICATE and edge[
'subject'] == disease_id and edge[
'object'] == target_id:
if 'properties' in edge.keys():
days = edge['properties']['days']
qualifier = edge['properties']['qualifier']
else:
days = 970
qualifier = '>='
total_edges += 1
total_nodes += 1
# set BKB target
dynamic_targets[node["id"]] = {
"op": qualifier,
"value": days,
}
truth_target = (node["id"], '{} {}'.format(qualifier, days))
# get evidence
for node_key in query["query_graph"]['nodes'].keys():
# genes
node = query["query_graph"]['nodes'][node_key]
if node['category'] == BIOLINK_GENE:
# check for appropriate gene node structure
gene_id = node_key
for edge_key in query["query_graph"]['edges'].keys():
edge = query["query_graph"]['edges'][edge_key]
if edge['predicate'] == BIOLINK_GENE_TO_DISEASE_PREDICATE and edge[
'subject'] == gene_id and edge[
'object'] == disease_id:
total_edges += 1
# check for appropriate gene node curie
gene_curie = node['id']
gene = gene_curie
evidence["_" + gene] = 'True'
total_nodes += 1
# drugs
if node['category'] == BIOLINK_DRUG:
# check for appropriate drug node structure
drug_id = node_key
for edge_key in query["query_graph"]['edges'].keys():
edge = query["query_graph"]['edges'][edge_key]
if edge['predicate'] == BIOLINK_CHEMICAL_TO_DISEASE_OR_PHENOTYPIC_FEATURE_PREDICATE and edge[
'subject'] == drug_id and edge[
'object'] == disease_id:
total_edges += 1
# check for appropriate drug node curie
drug_curie = node['id']
drug = drug_curie
evidence[node["id"]] = 'True'
total_nodes += 1
# produce BKB query
chp_query = Query(evidence=evidence,
targets=targets,
dynamic_evidence=dynamic_evidence,
dynamic_targets=dynamic_targets,
type='updating')
# Set some other helpful attributes
chp_query.truth_target = truth_target
chp_query.query_id = query["query_id"] if 'query_id' in query else None
return chp_query
def _run_query(self, chp_query, query_type):
if query_type == 'simple':
chp_query = self.joint_reasoner.run_query(chp_query)
# If a probability was found for the target
if len(chp_query.result) > 0:
# If a probability was found for the truth target
if chp_query.truth_target in chp_query.result:
total_unnormalized_prob = 0
for target, contrib in chp_query.result.items():
prob = max(0, contrib)
total_unnormalized_prob += prob
chp_query.truth_prob = max([
0, chp_query.result[(chp_query.truth_target)]
]) / total_unnormalized_prob
else:
chp_query.truth_prob = 0
else:
chp_query.truth_prob = -1
chp_query.report = None
else:
chp_query = self.dynamic_reasoner.run_query(chp_query)
chp_res_dict = chp_query.result.process_updates(normalize=True)
chp_query.truth_prob = max([
0, chp_res_dict[chp_query.truth_target[0]][
chp_query.truth_target[1]]
])
chp_query.report = None
return chp_query
def _construct_trapi_response(self, chp_query, query_type=None):
# Get orginal query
if len(self.init_query) == 1:
query = self.init_query[0]
query_id = None
else:
for _query in self.init_query:
if _query["query_id"] == chp_query.query_id:
query = _query
query_id = query["query_id"]
break
kg = copy.deepcopy(query["query_graph"])
# update target node info and form edge pair combos for results graph
node_pairs = dict()
for node_key in list(kg['nodes'].keys())[:]:
qg_node_curie = kg['nodes'][node_key].pop('id')
kg['nodes'][qg_node_curie] = kg['nodes'].pop(node_key)
node_pairs[node_key] = qg_node_curie
if kg['nodes'][qg_node_curie]['category'] == BIOLINK_GENE:
kg['nodes'][qg_node_curie]['name'] = self._get_curie_name(
BIOLINK_GENE, qg_node_curie)[0]
elif kg['nodes'][qg_node_curie]['category'] == BIOLINK_DRUG:
kg['nodes'][qg_node_curie]['name'] = self._get_curie_name(
BIOLINK_DRUG, qg_node_curie)[0]
edge_pairs = dict()
knowledge_edges = 0
for edge_key in list(kg['edges'].keys())[:]:
kg_id = 'kge{}'.format(knowledge_edges)
knowledge_edges += 1
kg['edges'][kg_id] = kg['edges'].pop(edge_key)
kg['edges'][kg_id]['subject'] = node_pairs[kg['edges'][kg_id]
['subject']]
kg['edges'][kg_id]['object'] = node_pairs[kg['edges'][kg_id]
['object']]
edge_pairs[edge_key] = kg_id
if kg['edges'][kg_id][
'predicate'] == BIOLINK_DISEASE_TO_PHENOTYPIC_FEATURE_PREDICATE:
if 'properties' in kg['edges'][kg_id].keys():
kg['edges'][kg_id].pop('properties')
kg['edges'][kg_id]['attributes'] = [{
'name':
'Probability of Survival',
'type':
BIOLINK_PROBABILITY,
'value':
chp_query.truth_prob
}]
results = []
results.append({
'edge_bindings': {},
'node_bindings': {},
})
for edge_pair_key in edge_pairs:
results[0]['edge_bindings'][edge_pair_key] = [{
'id':
edge_pairs[edge_pair_key]
}]
for node_pair_key in node_pairs:
results[0]['node_bindings'][node_pair_key] = [{
'id':
node_pairs[node_pair_key]
}]
# query response
trapi_message = {
'query_graph': query["query_graph"],
'knowledge_graph': kg,
'results': results
}
trapi_response = {'message': trapi_message}
return query_id, trapi_response
def setUpClass(cls):
super(TestBaseHandler, cls).setUpClass()
cls.bkb_handler = BkbDataHandler()
cls.dynamic_reasoner = ChpDynamicReasoner(cls.bkb_handler)
cls.joint_reasoner = ChpJointReasoner(cls.bkb_handler)
def setUpClass(cls):
super(TestDynamicReasoner, cls).setUpClass()
cls.bkb_handler = BkbDataHandler(disease='tcga_brca')
cls.dynamic_reasoner = ChpDynamicReasoner(cls.bkb_handler)
class WildCardHandlerMixin:
def _setup_handler(self):
# Only do the rest of this if a query is passed
if self.messages is not None:
# Setup messages
self._setup_messages()
# Instiatate Reasoners
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_messages(self):
if type(self.messages) == list:
self.message_dict = defaultdict(list)
for message in self.messages:
self.message_dict[self._get_wildcard_type(message)].append(message)
def _get_wildcard_type(self, message):
wildcard_type = None
for node_id, node in message.query_graph.nodes.items():
if node.ids is None:
if wildcard_type is None:
wildcard_type = node.categories[0]
if wildcard_type == BIOLINK_DRUG_ENTITY:
return 'drug'
elif wildcard_type == BIOLINK_GENE_ENTITY:
return 'gene'
else:
raise ValueError('Did not understand wildcard type {}.'.format(wildcard_type))
def _extract_chp_query(self, message, message_type):
# Initialize CHP BKB Query
chp_query = ChpQuery(reasoning_type='updating')
# ensure we are using all nodes/edges
total_nodes = 0
total_edges = 0
query_graph = message.query_graph
# get phenotype node
targets = list()
acceptable_target_curies = ['EFO:0000714']
self.implicit_survival_node = False
for node_key in query_graph.nodes.keys():
node = query_graph.nodes[node_key]
if node.categories[0] == BIOLINK_PHENOTYPIC_FEATURE_ENTITY and node.ids[0] in acceptable_target_curies:
target_id = node_key
total_nodes += 1
if total_nodes == 0:
# Use Default Survival
self.implicit_survival_node = True
total_nodes += 1
#acceptable_target_curies_print = ','.join(acceptable_target_curies)
#sys.exit("Survival Node not found. Node category must be '{}' and id must be in: {}".format(Biolink(BIOLINK_PHENOTYPIC_FEATURE),
# acceptable_target_curies_print))
survival_value = 970
survival_operator = '>='
# get disease node info and ensure only 1 disease:
acceptable_disease_curies = ['MONDO:0007254']
for node_key in query_graph.nodes.keys():
node = query_graph.nodes[node_key]
if node.categories[0] == BIOLINK_DISEASE_ENTITY and node.ids[0] in acceptable_disease_curies:
disease_id = node_key
for edge_key in query_graph.edges.keys():
edge = query_graph.edges[edge_key]
if self.check_predicate_support(edge.predicates[0], BIOLINK_HAS_PHENOTYPE_ENTITY) and edge.subject == disease_id and edge.object == target_id:
survival_time_constraint = edge.find_constraint(name='survival_time')
if survival_time_constraint is not None:
survival_value = survival_time_constraint.value
survival_operator = survival_time_constraint.operator
if survival_operator == 'matches':
survival_operator = '=='
total_edges += 1
total_nodes += 1
if self.implicit_survival_node:
days=970
qualifier = '>='
total_edges += 1
# set BKB target
chp_query.add_dynamic_target('EFO:0000714', survival_operator, survival_value)
truth_target = ('EFO:0000714', '{} {}'.format(survival_operator, survival_value))
# get evidence
for node_key in query_graph.nodes.keys():
# genes
node = query_graph.nodes[node_key]
if node.categories[0] == BIOLINK_GENE_ENTITY:
# check for appropriate gene node structure
gene_id = node_key
for edge_key in query_graph.edges.keys():
edge = query_graph.edges[edge_key]
if self.check_predicate_support(edge.predicates[0], BIOLINK_GENE_ASSOCIATED_WITH_CONDITION_ENTITY) and edge.subject == gene_id and edge.object == disease_id:
total_edges += 1
# check for appropriate gene node curie
if message_type != 'gene':
gene_curie = node.ids[0]
if gene_curie in self.curies[BIOLINK_GENE_ENTITY.get_curie()]:
gene = gene_curie
chp_query.add_meta_evidence(gene, 'True')
total_nodes += 1
# drugs
if node.categories[0] == BIOLINK_DRUG_ENTITY:
# check for appropriate drug node structure
drug_id = node_key
for edge_key in query_graph.edges.keys():
edge = query_graph.edges[edge_key]
if self.check_predicate_support(edge.predicates[0], BIOLINK_TREATS_ENTITY) and edge.subject == drug_id and edge.object == disease_id:
total_edges += 1
# check for appropriate drug node curie
if message_type != 'drug':
drug_curie = node.ids[0]
if drug_curie in self.curies[BIOLINK_DRUG_ENTITY.get_curie()]:
drug = drug_curie
chp_query.add_meta_evidence(drug, 'True')
total_nodes += 1
# Temporary solution to no evidence linking
if len(chp_query.evidence.keys()) == 0 and len(chp_query.dynamic_evidence.keys()) == 0:
self.no_evidence_probability_check = True
else:
self.no_evidence_probability_check = False
# Set some other helpful attributes
chp_query.truth_target = truth_target
return chp_query
def _run_query(self, chp_query, query_type):
""" Runs build BKB query to calculate probability of survival.
A probability is returned to specificy survival time w.r.t a drug.
Contributions for each gene are calculuated and classified under
their true/false target assignments.
"""
# temporary solution to no evidence linking
if not self.no_evidence_probability_check:
if query_type == 'gene':
chp_query = self.dynamic_reasoner.run_query(chp_query, bkb_type='drug')
elif query_type == 'drug':
chp_query = self.dynamic_reasoner.run_query(chp_query, bkb_type='gene')
chp_res_dict = chp_query.result.process_updates()
chp_res_norm_dict = chp_query.result.process_updates(normalize=True)
#chp_query.result.summary()
chp_res_contributions = chp_query.result.process_inode_contributions()
try:
chp_query.truth_prob = max([0, chp_res_dict[chp_query.truth_target[0]][chp_query.truth_target[1]]])
except KeyError:
# May need to come back and fix this.
chp_query.truth_prob = -1
# Collect all source inodes and process patient hashes
patient_contributions = defaultdict(lambda: defaultdict(int))
for target, contrib_dict in chp_res_contributions.items():
target_comp_name, target_state_name = target
for inode, contrib in contrib_dict.items():
comp_name, state_name = inode
if '_Source_' in comp_name:
# Split source state name to get patient hashes
source_hashes_str = state_name.split('_')[-1]
source_hashes = [int(source_hash) for source_hash in source_hashes_str.split(',')]
hash_len = len(source_hashes)
# Process patient contributions
for _hash in source_hashes:
# Normalize to get relative contribution
patient_contributions[target][_hash] += contrib/hash_len #/ chp_res_dict[target_comp_name][target_state_name]
else:
# probability of survival
num_survived = 0
num_all = len(self.dynamic_reasoner.raw_patient_data.keys())
str_op = chp_query.dynamic_targets['EFO:0000714']['op']
opp_op = get_opposite_operator(str_op)
op = get_operator(str_op)
days = chp_query.dynamic_targets['EFO:0000714']['value']
for patient, pat_dict in self.dynamic_reasoner.raw_patient_data.items():
if op(pat_dict['survival_time'], days):
num_survived += 1
chp_query.truth_prob = num_survived/num_all
# patient_contributions
patient_contributions = defaultdict(lambda: defaultdict(int))
for patient, pat_dict in self.dynamic_reasoner.raw_patient_data.items():
if op(pat_dict['survival_time'], days):
if num_survived == 0:
patient_contributions[('EFO:0000714', '{} {}'.format(str_op, days))][patient] = 0
else:
patient_contributions[('EFO:0000714', '{} {}'.format(str_op, days))][patient] = chp_query.truth_prob/num_survived
else:
if num_survived == 0:
patient_contributions[('EFO:0000714', '{} {}'.format(opp_op, days))][patient] = (1-chp_query.truth_prob)/num_all
else:
patient_contributions[('EFO:0000714', '{} {}'.format(opp_op, days))][patient] = (1-chp_query.truth_prob)/(num_all-num_survived)
# Now iterate through the patient data to translate patient contributions to drug/gene contributions
wildcard_contributions = defaultdict(lambda: defaultdict(int))
for target, patient_contrib_dict in patient_contributions.items():
for patient, contrib in patient_contrib_dict.items():
if query_type == 'gene':
for gene_curie in self.dynamic_reasoner.raw_patient_data[patient]["gene_curies"]:
wildcard_contributions[gene_curie][target] += contrib
elif query_type == 'drug':
for drug_curie in self.dynamic_reasoner.raw_patient_data[patient]["drug_curies"]:
wildcard_contributions[drug_curie][target] += contrib
# normalize gene contributions by the target and take relative difference
for curie in wildcard_contributions.keys():
truth_target_gene_contrib = 0
nontruth_target_gene_contrib = 0
for target, contrib in wildcard_contributions[curie].items():
if target[0] == chp_query.truth_target[0] and target[1] == chp_query.truth_target[1]:
truth_target_gene_contrib += contrib / chp_query.truth_prob
else:
nontruth_target_gene_contrib += contrib / (1 - chp_query.truth_prob)
wildcard_contributions[curie]['relative'] = truth_target_gene_contrib - nontruth_target_gene_contrib
chp_query.report = None
chp_query.wildcard_contributions = wildcard_contributions
return chp_query
def _construct_trapi_message(self, chp_query, message, query_type=None):
# update target node info and form edge pair combos for results graph
qg = message.query_graph
kg = message.knowledge_graph
# Process Standard QUery as first result.
# Process Nodes
node_bindings = {}
contrib_qg_id = None
for qnode_key, qnode in qg.nodes.items():
if qnode.ids is not None:
if qnode.categories[0] == BIOLINK_GENE_ENTITY:
knode_key = kg.add_node(
qnode.ids[0],
self.curies[BIOLINK_GENE_ENTITY.get_curie()][qnode.ids[0]][0],
qnode.categories[0].get_curie(),
)
elif qnode.categories[0] == BIOLINK_DRUG_ENTITY:
knode_key = kg.add_node(
qnode.ids[0],
self.curies[BIOLINK_DRUG_ENTITY.get_curie()][qnode.ids[0]][0],
qnode.categories[0].get_curie(),
)
else:
knode_key = kg.add_node(
qnode.ids[0],
qnode.ids[0],
qnode.categories[0].get_curie(),
)
node_bindings[qnode_key] = [knode_key]
if not self.implicit_survival_node:
# Process Edges
edge_bindings = {}
knowledge_edges = 0
for qedge_key, qedge in qg.edges.items():
if not qedge.subject in node_bindings or not qedge.object in node_bindings:
continue
kedge_key = kg.add_edge(
node_bindings[qedge.subject][0],
node_bindings[qedge.object][0],
predicate=qedge.predicates[0].get_curie(),
relation=qedge.relation,
)
edge_bindings[qedge_key] = [kedge_key]
# Add Attribute
if self.check_predicate_support(qedge.predicates[0], BIOLINK_HAS_PHENOTYPE_ENTITY):
kg.edges[kedge_key].add_attribute(
attribute_type_id='Probability of Survival',
value=chp_query.truth_prob,
value_type_id=BIOLINK_HAS_CONFIDENCE_LEVEL_ENTITY.get_curie(),
)
'''
subject_node = kg['edges'][edge_key]['subject']
if kg['edges'][edge_key]['predicate'] == BIOLINK_GENE_ENTITY_TO_DISEASE_PREDICATE, is_slot=True) and query['query_graph']['nodes'][subject_node]['category'] == BIOLINK_GENE_ENTITY and query_type == 'gene':
kg['edges'].pop(edge_key)
elif kg['edges'][edge_key]['predicate'] == BIOLINK_CHEMICAL_TO_DISEASE_OR_PHENOTYPIC_FEATURE_PREDICATE, is_slot=True) and query['query_graph']['nodes'][subject_node]['category'] == BIOLINK_DRUG_ENTITY and query_type == 'drug':
kg['edges'].pop(edge_key)
else:
kg_id = 'kge{}'.format(knowledge_edges)
knowledge_edges += 1
kg['edges'][kg_id] = kg['edges'].pop(edge_key)
kg['edges'][kg_id]['subject'] = node_pairs[kg['edges'][kg_id]['subject']]
kg['edges'][kg_id]['object'] = node_pairs[kg['edges'][kg_id]['object']]
edge_pairs[edge_key] = kg_id
if kg['edges'][kg_id]['predicate'] == BIOLINK_DISEASE_ENTITY_TO_PHENOTYPIC_FEATURE_PREDICATE, is_slot=True):
if 'properties' in kg['edges'][kg_id].keys():
kg['edges'][kg_id].pop('properties')
kg['edges'][kg_id]['attributes'] = [{'name':'Probability of Survival',
'type':BIOLINK_PROBABILITY,
'value':chp_query.truth_prob}]
'''
# Proces results
message.results.add_result(
node_bindings,
edge_bindings,
)
'''
# Put first result of standard prob query of only curie nodes (i.e. no wildcard nodes where used as evidence)
results = []
results.append({'edge_bindings':dict(),
'node_bindings':dict()})
for edge_pair_key in edge_pairs:
results[0]['edge_bindings'][edge_pair_key] = [{ 'id': str(edge_pairs[edge_pair_key])}]
for node_pair_key in node_pairs:
results[0]['node_bindings'][node_pair_key] = [{ 'id': str(node_pairs[node_pair_key])}]
'''
#else:
# knowledge_edges = 0
# kg['edges'] = {}
# results = []
# Build relative contribution results and added associated edges into knowledge graph
unsorted_wildcard_contributions = []
for wildcard, contrib_dict in chp_query.wildcard_contributions.items():
unsorted_wildcard_contributions.append((contrib_dict['relative'], wildcard))
sorted_wildcard_contributions = [(contrib,wildcard) for contrib, wildcard in sorted(unsorted_wildcard_contributions, key=lambda x: abs(x[0]), reverse=True)]
for contrib, wildcard in sorted_wildcard_contributions[:self.max_results]:
#TODO: Fix this!
if wildcard == 'missing':
continue
#rg = copy.deepcopy(query["query_graph"])
_node_bindings = {}
_edge_bindings = {}
# Process node bindings
bad_wildcard = False
for qnode_id, qnode in qg.nodes.items():
if qnode.categories[0] == BIOLINK_GENE_ENTITY and query_type == 'gene':
try:
knode_id = kg.add_node(
wildcard,
self.curies[BIOLINK_GENE_ENTITY.get_curie()][wildcard][0],
qnode.categories[0].get_curie(),
)
_node_bindings[qnode_id] = [knode_id]
except KeyError:
logger.info("Couldn't find {} in curies[{}]".format(wildcard, BIOLINK_GENE_ENTITY.get_curie()))
bad_wildcard = True
elif qnode.categories[0] == BIOLINK_DRUG_ENTITY and query_type == 'drug':
knode_id = kg.add_node(
wildcard,
self.curies[BIOLINK_DRUG_ENTITY.get_curie()][wildcard][0],
qnode.categories[0].get_curie(),
)
_node_bindings[qnode_id] = [knode_id]
else:
_node_bindings[qnode_id] = node_bindings[qnode_id]
if bad_wildcard:
continue
# Process edge bindings
for qedge_id, qedge in qg.edges.items():
subject_node = qedge.subject
object_node = qedge.object
if query_type == 'gene' and self.check_predicate_support(qedge.predicates[0], BIOLINK_GENE_ASSOCIATED_WITH_CONDITION_ENTITY) and qg.nodes[subject_node].categories[0] == BIOLINK_GENE_ENTITY:
kedge_id = kg.add_edge(
_node_bindings[qedge.subject][0],
_node_bindings[qedge.object][0],
predicate=qedge.predicates[0],
relation=qedge.relation,
)
kg.edges[kedge_id].add_attribute(
attribute_type_id='Contribution',
value=contrib,
value_type_id=BIOLINK_HAS_EVIDENCE_ENTITY.get_curie(),
)
_edge_bindings[qedge_id] = [kedge_id]
elif query_type == 'gene' and self.check_predicate_support(qedge.predicates[0], BIOLINK_CONDITION_ASSOCIATED_WITH_GENE_ENTITY) and qg.nodes[object_node].categories[0] == BIOLINK_GENE_ENTITY:
kedge_id = kg.add_edge(
_node_bindings[qedge.subject][0],
_node_bindings[qedge.object][0],
predicate=qedge.predicates[0],
relation=qedge.relation,
)
kg.edges[kedge_id].add_attribute(
attribute_type_id='Contribution',
value=contrib,
value_type_id=BIOLINK_HAS_EVIDENCE_ENTITY.get_curie(),
)
_edge_bindings[qedge_id] = [kedge_id]
elif query_type == 'drug' and self.check_predicate_support(qedge.predicates[0], BIOLINK_TREATS_ENTITY) and qg.nodes[subject_node].categories[0] == BIOLINK_DRUG_ENTITY:
kedge_id = kg.add_edge(
_node_bindings[qedge.subject][0],
_node_bindings[qedge.object][0],
predicate=qedge.predicates[0],
relation=qedge.relation,
)
kg.edges[kedge_id].add_attribute(
attribute_type_id='Contribution',
value=contrib,
value_type_id=BIOLINK_HAS_EVIDENCE_ENTITY.get_curie(),
)
_edge_bindings[qedge_id] = [kedge_id]
elif query_type == 'drug' and self.check_predicate_support(qedge.predicates[0], BIOLINK_TREATED_BY_ENTITY) and qg.nodes[object_node].categories[0] == BIOLINK_DRUG_ENTITY:
kedge_id = kg.add_edge(
_node_bindings[qedge.subject][0],
_node_bindings[qedge.object][0],
predicate=qedge.predicates[0],
relation=qedge.relation,
)
kg.edges[kedge_id].add_attribute(
attribute_type_id='Contribution',
value=contrib,
value_type_id=BIOLINK_HAS_EVIDENCE_ENTITY.get_curie(),
)
_edge_bindings[qedge_id] = [kedge_id]
else:
_edge_bindings[qedge_id] = edge_bindings[qedge_id]
# Process node and edge binding results
message.results.add_result(
_node_bindings,
_edge_bindings,
)
return message
class OneHopHandlerMixin:
""" OneHopeHandler is the handler for 1-hop queries. That is
query graphs (QGs) that consists of 2 nodes and a single edge.
:param query: the query graph sent by the ARA.
:type query: dict
:param hosts_filename: a filename for a stored QG. Defaults to None
:type hosts_filename: str
:param num_processes_per_host: Not implemented thouroughly, but would be
used for distributed reasoning.
:type num_processes_per_host: int
:param max_results: specific to 1-hop queries, specifies the number of
wildcard genes to return.
:type max_results: int
"""
def _setup_handler(self):
self.default_survival_target = {
"EFO:0000714": {
"op": '>=',
"value": 970
}
}
# Only do the rest of this if a query is passed
if self.messages is not None:
# Setup queries
self._setup_messages()
# Instiatate Reasoners
if self.dynamic_reasoner is None:
self.dynamic_reasoner = ChpDynamicReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
if self.joint_reasoner is None:
self.joint_reasoner = ChpJointReasoner(
bkb_handler=self.bkb_data_handler,
hosts_filename=self.hosts_filename,
num_processes_per_host=self.num_processes_per_host)
def _setup_messages(self):
self.message_dict = defaultdict(list)
for message in self.messages:
self.message_dict[self._get_onehop_type(message)].append(message)
def _get_onehop_type(self, message):
wildcard_type = None
for node_id, node in message.query_graph.nodes.items():
if node.ids is None:
if wildcard_type is None:
wildcard_type = node.categories[0]
# If standard onehop query
if wildcard_type is None:
return 'standard'
elif wildcard_type == BIOLINK_DRUG_ENTITY:
return 'drug'
elif wildcard_type == BIOLINK_GENE_ENTITY:
return 'gene'
else:
raise ValueError(
'Did not understand wildcard type {}.'.format(wildcard_type))
def check_query(self):
""" Currently not implemented. Would check validity of query.
"""
return True
@staticmethod
def _process_predicate_proxy(qedge):
dynamic_targets = {}
predicate_proxy_constraint = qedge.find_constraint('predicate_proxy')
if predicate_proxy_constraint is None:
predicate_proxy = get_default_predicate_proxy()
proxy_constraint = qedge.find_constraint(predicate_proxy)
else:
predicate_proxy = predicate_proxy_constraint.value[0]
proxy_constraint = qedge.find_constraint(predicate_proxy)
if proxy_constraint is None:
proxy_operator = get_default_operator(predicate_proxy)
proxy_value = get_default_value(predicate_proxy)
else:
proxy_operator = proxy_constraint.operator
proxy_value = proxy_constraint.value
# Setup dynamic target
dynamic_targets[predicate_proxy] = {
"op": proxy_operator,
"value": proxy_value,
}
return dynamic_targets
@staticmethod
def _process_predicate_context(qedge, message_type):
evidence = {}
dynamic_evidence = {}
predicate_context_constraint = qedge.find_constraint(
'predicate_context')
if predicate_context_constraint is not None:
for context in predicate_context_constraint.value:
context_curie = get_biolink_entity(context)
context_constraint = qedge.find_constraint(context)
if context_constraint is None:
raise ValueError(
'Provided no context details for {}'.format(context))
if context_curie == BIOLINK_GENE_ENTITY:
if message_type == 'gene':
if type(context_constraint.value) is list:
for _curie in context_constraint.value:
dynamic_evidence[_curie] = {
"op": '==',
"value": 'True',
}
else:
dynamic_evidence[context_constraint.value] = {
"op": '==',
"value": 'True',
}
else:
if type(context_constraint.value) is list:
for _curie in context_constraint.value:
evidence['_{}'.format(_curie)] = 'True'
else:
evidence['_{}'.format(_curie)] = 'True'
elif context_curie == BIOLINK_DRUG_ENTITY:
if message_type == 'drug':
if type(context_constraint.value) is list:
for _curie in context_constraint.value:
dynamic_evidence[_curie] = {
"op": '==',
"value": 'True',
}
else:
dynamic_evidence[context_constraint.value] = {
"op": '==',
"value": 'True',
}
else:
if type(context_constraint.value) is list:
for _curie in context_constraint.value:
evidence['_{}'.format(_curie)] = 'True'
else:
evidence['_{}'.format(_curie)] = 'True'
else:
raise ValueError(
'Unsupported context type: {}'.format(context_curie))
return evidence, dynamic_evidence
def _extract_chp_query(self, message, message_type):
evidence = {}
dynamic_targets = {}
dynamic_evidence = {}
if message_type == 'standard':
# Setup gene and drug evidence
for qnode_id, qnode in message.query_graph.nodes.items():
if qnode.categories[
0] == BIOLINK_GENE_ENTITY or qnode.categories[
0] == BIOLINK_DRUG_ENTITY:
evidence['_{}'.format(qnode.ids[0])] = 'True'
elif message_type == 'gene':
for qnode_id, qnode in message.query_graph.nodes.items():
if qnode.categories[0] == BIOLINK_DRUG_ENTITY:
#dynamic_evidence[qnode.ids[0]] = {
# "op": '==',
# "value": 'True',
# }
evidence['_{}'.format(qnode.ids[0])] = 'True'
elif message_type == 'drug':
for qnode_id, qnode in message.query_graph.nodes.items():
if qnode.categories[0] == BIOLINK_GENE_ENTITY:
#dynamic_evidence[qnode.ids[0]] = {
# "op": '==',
# "value": 'True',
# }
evidence['_{}'.format(qnode.ids[0])] = 'True'
# Grab edge
for qedge_id, qedge in message.query_graph.edges.items():
break
# Process predicate proxy
dynamic_targets = self._process_predicate_proxy(qedge)
# Process predicate context
_evidence, _dynamic_evidence = self._process_predicate_context(
qedge, message_type)
evidence.update(_evidence)
dynamic_evidence.update(_dynamic_evidence)
#TODO: Probably need a more robust solution for when no context is provided in wildcard queries and you need it.
#if len(evidence) == 0:
# raise ValueError('Did not supply context with a query that required context.')
target = list(dynamic_targets.keys())[0]
truth_target = (target,
'{} {}'.format(dynamic_targets[target]["op"],
dynamic_targets[target]["value"]))
chp_query = Query(evidence=evidence,
targets=None,
dynamic_evidence=dynamic_evidence,
dynamic_targets=dynamic_targets,
type='updating')
# Set some other helpful attributes
chp_query.truth_target = truth_target
return chp_query
def _run_query(self, chp_query, query_type):
""" Runs build BKB query to calculate probability of survival.
A probability is returned to specificy survival time w.r.t a drug.
Contributions for each gene are calculuated and classified under
their true/false target assignments.
"""
if query_type == 'standard':
chp_query = self.joint_reasoner.run_query(chp_query)
# If a probability was found for the target
if len(chp_query.result) > 0:
# If a probability was found for the truth target
if chp_query.truth_target in chp_query.result:
total_unnormalized_prob = 0
for target, contrib in chp_query.result.items():
prob = max(0, contrib)
total_unnormalized_prob += prob
chp_query.truth_prob = max([
0, chp_query.result[(chp_query.truth_target)]
]) / total_unnormalized_prob
else:
chp_query.truth_prob = 0
else:
chp_query.truth_prob = -1
chp_query.report = None
return chp_query
else:
# Do this if a disease node is present
if len(chp_query.evidence) == 0:
# probability of survival
chp_query = self.joint_reasoner.run_query(chp_query)
if len(chp_query.result) > 0:
# If a probability was found for the truth target
if chp_query.truth_target in chp_query.result:
total_unnormalized_prob = 0
for target, contrib in chp_query.result.items():
prob = max(0, contrib)
total_unnormalized_prob += prob
chp_query.truth_prob = max([
0, chp_query.result[(chp_query.truth_target)]
]) / total_unnormalized_prob
else:
chp_query.truth_prob = 0
else:
chp_query.truth_prob = -1
# patient_contributions
num_all = len(self.joint_reasoner.patient_data)
num_matched = chp_query.truth_prob * num_all
patient_contributions = defaultdict(lambda: defaultdict(int))
for patient, feature_dict in self.joint_reasoner.patient_data.items(
):
for predicate_proxy, proxy_info in chp_query.dynamic_targets.items(
):
proxy_op = get_operator(proxy_info["op"])
proxy_opp_op = get_opposite_operator(proxy_info["op"])
proxy_value = proxy_info["value"]
if proxy_op(feature_dict[predicate_proxy],
proxy_value):
if num_matched == 0:
patient_contributions[(
predicate_proxy,
'{} {}'.format(proxy_op,
proxy_value))][patient] = 0
else:
patient_contributions[(
predicate_proxy,
'{} {}'.format(proxy_op, proxy_value)
)][patient] = chp_query.truth_prob / num_matched
else:
if num_matched == 0:
patient_contributions[(
predicate_proxy,
'{} {}'.format(proxy_opp_op, proxy_value)
)][patient] = (
1 - chp_query.truth_prob) / num_matched
else:
patient_contributions[(
predicate_proxy,
'{} {}'.format(proxy_opp_op, proxy_value)
)][patient] = (1 - chp_query.truth_prob) / (
num_all - num_matched)
'''
num_survived = 0
num_all = len(self.dynamic_reasoner.raw_patient_data.keys())
str_op = chp_query.dynamic_targets['EFO:0000714']['op']
opp_op = get_opposite_operator(str_op)
op = get_operator(str_op)
days = chp_query.dynamic_targets['EFO:0000714']['value']
for patient, pat_dict in self.dynamic_reasoner.raw_patient_data.items():
if op(pat_dict['survival_time'], days):
num_survived += 1
chp_query.truth_prob = num_survived/num_all
# patient_contributions
patient_contributions = defaultdict(lambda: defaultdict(int))
for patient, pat_dict in self.dynamic_reasoner.raw_patient_data.items():
if op(pat_dict['survival_time'], days):
if num_survived == 0:
patient_contributions[('EFO:0000714', '{} {}'.format(str_op, days))][patient] = 0
else:
patient_contributions[('EFO:0000714', '{} {}'.format(str_op, days))][patient] = chp_query.truth_prob/num_survived
else:
if num_survived == 0:
patient_contributions[('EFO:0000714', '{} {}'.format(opp_op, days))][patient] = (1-chp_query.truth_prob)/num_all
else:
patient_contributions[('EFO:0000714', '{} {}'.format(opp_op, days))][patient] = (1-chp_query.truth_prob)/(num_all-num_survived)
'''
else:
if query_type == 'gene':
chp_query = self.dynamic_reasoner.run_query(
chp_query, bkb_type='drug')
elif query_type == 'drug':
chp_query = self.dynamic_reasoner.run_query(
chp_query, bkb_type='gene')
chp_res_dict = chp_query.result.process_updates()
chp_res_norm_dict = chp_query.result.process_updates(
normalize=True)
#chp_query.result.summary()
chp_res_contributions = chp_query.result.process_inode_contributions(
)
chp_query.truth_prob = max([
0, chp_res_norm_dict[chp_query.truth_target[0]][
chp_query.truth_target[1]]
])
# Collect all source inodes and process patient hashes
patient_contributions = defaultdict(lambda: defaultdict(int))
for target, contrib_dict in chp_res_contributions.items():
target_comp_name, target_state_name = target
for inode, contrib in contrib_dict.items():
comp_name, state_name = inode
if '_Source_' in comp_name:
# Split source state name to get patient hashes
source_hashes_str = state_name.split('_')[-1]
source_hashes = [
int(source_hash)
for source_hash in source_hashes_str.split(',')
]
hash_len = len(source_hashes)
# Process patient contributions
for _hash in source_hashes:
# Normalize to get relative contribution
patient_contributions[target][
_hash] += contrib / hash_len #/ chp_res_dict[target_comp_name][target_state_name]
# Now iterate through the patient data to translate patient contributions to drug/gene contributions
wildcard_contributions = defaultdict(lambda: defaultdict(int))
for target, patient_contrib_dict in patient_contributions.items():
for patient, contrib in patient_contrib_dict.items():
if query_type == 'gene':
for gene_curie in self.dynamic_reasoner.raw_patient_data[
int(patient)]["gene_curies"]:
wildcard_contributions[gene_curie][target] += contrib
elif query_type == 'drug':
for drug_curie in self.dynamic_reasoner.raw_patient_data[
int(patient)]["drug_curies"]:
wildcard_contributions[drug_curie][target] += contrib
# normalize gene contributions by the target and take relative difference
for curie in wildcard_contributions.keys():
truth_target_gene_contrib = 0
nontruth_target_gene_contrib = 0
for target, contrib in wildcard_contributions[curie].items():
if target[0] == chp_query.truth_target[0] and target[
1] == chp_query.truth_target[1]:
truth_target_gene_contrib += contrib / chp_query.truth_prob
else:
nontruth_target_gene_contrib += contrib / (
1 - chp_query.truth_prob)
wildcard_contributions[curie][
'relative'] = truth_target_gene_contrib - nontruth_target_gene_contrib
chp_query.report = None
chp_query.wildcard_contributions = wildcard_contributions
return chp_query
def _construct_trapi_message(self, chp_query, message, query_type):
qg = message.query_graph
kg = message.knowledge_graph
edge_bindings = {}
node_bindings = {}
# Process nodes
for qnode_id, qnode in qg.nodes.items():
if qnode.ids is not None:
if qnode.categories[0] == BIOLINK_GENE_ENTITY:
knode_key = kg.add_node(
qnode.ids[0],
self.curies[BIOLINK_GENE_ENTITY.get_curie()][
qnode.ids[0]][0],
qnode.categories[0].get_curie(),
)
elif qnode.categories[0] == BIOLINK_DRUG_ENTITY:
knode_key = kg.add_node(
qnode.ids[0],
self.curies[BIOLINK_DRUG_ENTITY.get_curie()][
qnode.ids[0]][0],
qnode.categories[0].get_curie(),
)
elif qnode.categories[0] == BIOLINK_DISEASE_ENTITY:
#TODO: Add diseases to curies and fix name hack below.
knode_key = kg.add_node(
qnode.ids[0],
qnode.
ids[0], #TODO: Once curies is fixed, make this a name.
qnode.categories[0].get_curie(),
)
node_bindings[qnode_id] = [knode_key]
else:
wildcard_node = qnode
if query_type == 'standard':
for qedge_key, qedge in qg.edges.items():
kedge_key = kg.add_edge(
node_bindings[qedge.subject][0],
node_bindings[qedge.object][0],
predicate=qedge.predicates[0].get_curie(),
relation=qedge.relation,
)
edge_bindings[qedge_key] = [kedge_key]
# Add Attribute
kg.edges[kedge_key].add_attribute(
attribute_type_id='Probability of Survival',
value=chp_query.truth_prob,
value_type_id=BIOLINK_HAS_CONFIDENCE_LEVEL_ENTITY.
get_curie(),
)
message.results.add_result(
node_bindings,
edge_bindings,
)
else:
# Build relative contribution results and added associated edges into knowledge graph
unsorted_wildcard_contributions = []
for wildcard, contrib_dict in chp_query.wildcard_contributions.items(
):
unsorted_wildcard_contributions.append(
(contrib_dict['relative'], wildcard))
sorted_wildcard_contributions = [
(contrib, wildcard) for contrib, wildcard in sorted(
unsorted_wildcard_contributions,
key=lambda x: abs(x[0]),
reverse=True)
]
# add kg gene nodes and edges
edge_count = 0
node_count = 1
results = []
for contrib, wildcard in sorted_wildcard_contributions[:self.
max_results]:
_node_bindings = {}
_edge_bindings = {}
# Process node bindings
bad_wildcard = False
for qnode_id, qnode in qg.nodes.items():
if qnode.categories[
0] == BIOLINK_GENE_ENTITY and query_type == 'gene':
try:
knode_id = kg.add_node(
wildcard,
self.curies[BIOLINK_GENE_ENTITY.get_curie()]
[wildcard][0],
qnode.categories[0].get_curie(),
)
_node_bindings[qnode_id] = [knode_id]
except KeyError:
logger.info(
"Couldn't find {} in curies[{}]".format(
wildcard, BIOLINK_GENE))
bad_wildcard = True
elif qnode.categories[
0] == BIOLINK_DRUG_ENTITY and query_type == 'drug':
knode_id = kg.add_node(
wildcard,
self.curies[BIOLINK_DRUG_ENTITY.get_curie()]
[wildcard][0],
qnode.categories[0].get_curie(),
)
_node_bindings[qnode_id] = [knode_id]
else:
_node_bindings[qnode_id] = node_bindings[qnode_id]
if bad_wildcard:
continue
# Process edge bindings
for qedge_id, qedge in qg.edges.items():
kedge_id = kg.add_edge(
_node_bindings[qedge.subject][0],
_node_bindings[qedge.object][0],
predicate=qedge.predicates[0],
relation=qedge.relation,
)
kg.edges[kedge_id].add_attribute(
attribute_type_id='Contribution',
value=contrib,
value_type_id=BIOLINK_HAS_EVIDENCE_ENTITY.get_curie(),
)
_edge_bindings[qedge_id] = [kedge_id]
# Process node and edge binding results
message.results.add_result(
_node_bindings,
_edge_bindings,
)
return message