def _add_component_pathway_association(self, gene_curie, pathway_curie,
pathway_label, eco_curie):
pathway = Pathway(self.graph)
pathway.addPathway(pathway_curie, pathway_label)
pathway.addComponentToPathway(gene_curie, pathway_curie)
association = Assoc(self.graph, self.name)
association.sub = gene_curie
association.rel = self.globaltt['involved in']
association.obj = pathway_curie
association.set_association_id()
association.add_evidence(eco_curie)
association.add_association_to_graph()
def _process_pathways(self, limit=None):
"""
This method adds the KEGG pathway IDs.
These are the canonical pathways as defined in KEGG.
We also encode the graphical depiction
which maps 1:1 with the identifier.
Triples created:
<pathway_id> is a GO:signal_transduction
<pathway_id> rdfs:label <pathway_name>
<gene_id> RO:involved_in <pathway_id>
:param limit:
:return:
"""
logger.info("Processing pathways")
if self.testMode:
g = self.testgraph
else:
g = self.graph
line_counter = 0
path = Pathway(g, self.nobnodes)
gu = GraphUtils(curie_map.get())
raw = '/'.join((self.rawdir, self.files['pathway']['file']))
with open(raw, 'r', encoding="iso-8859-1") as csvfile:
filereader = csv.reader(csvfile, delimiter='\t', quotechar='\"')
for row in filereader:
line_counter += 1
(pathway_id, pathway_name) = row
if self.testMode and \
pathway_id not in self.test_ids['pathway']:
continue
pathway_id = 'KEGG-'+pathway_id.strip()
path.addPathway(pathway_id, pathway_name)
# we know that the pathway images from kegg map 1:1 here.
# so add those
image_filename = re.sub(r'KEGG-path:', '', pathway_id) + '.png'
image_url = \
'http://www.genome.jp/kegg/pathway/map/'+image_filename
gu.addDepiction(g, pathway_id, image_url)
if not self.testMode and \
limit is not None and line_counter > limit:
break
logger.info("Done with pathways")
return
def _process_pathways(self, limit=None):
"""
This method adds the KEGG pathway IDs.
These are the canonical pathways as defined in KEGG.
We also encode the graphical depiction
which maps 1:1 with the identifier.
Triples created:
<pathway_id> is a GO:signal_transduction
<pathway_id> rdfs:label <pathway_name>
<gene_id> RO:involved_in <pathway_id>
:param limit:
:return:
"""
logger.info("Processing pathways")
if self.testMode:
g = self.testgraph
else:
g = self.graph
model = Model(g)
line_counter = 0
path = Pathway(g)
raw = '/'.join((self.rawdir, self.files['pathway']['file']))
with open(raw, 'r', encoding="iso-8859-1") as csvfile:
filereader = csv.reader(csvfile, delimiter='\t', quotechar='\"')
for row in filereader:
line_counter += 1
(pathway_id, pathway_name) = row
if self.testMode and \
pathway_id not in self.test_ids['pathway']:
continue
pathway_id = 'KEGG-'+pathway_id.strip()
path.addPathway(pathway_id, pathway_name)
# we know that the pathway images from kegg map 1:1 here.
# so add those
image_filename = re.sub(r'KEGG-path:', '', pathway_id) + '.png'
image_url = \
'http://www.genome.jp/kegg/pathway/map/'+image_filename
model.addDepiction(pathway_id, image_url)
if not self.testMode and \
limit is not None and line_counter > limit:
break
logger.info("Done with pathways")
return
def _add_component_pathway_association(
self, eco_map, component, component_prefix, pathway_id,
pathway_prefix, pathway_label, go_ecode):
pathway = Pathway(self.graph)
pathway_curie = "{}:{}".format(pathway_prefix, pathway_id)
gene_curie = "{}:{}".format(component_prefix, component.strip())
eco_curie = eco_map[go_ecode]
pathway.addPathway(pathway_curie, pathway_label)
pathway.addComponentToPathway(gene_curie, pathway_curie)
association = Assoc(self.graph, self.name)
association.sub = gene_curie
association.rel = pathway.object_properties['involved_in']
association.obj = pathway_curie
association.set_association_id()
association.add_evidence(eco_curie)
association.add_association_to_graph()
return
def _add_component_pathway_association(
self, eco_map, component, component_prefix, pathway_id,
pathway_prefix, pathway_label, go_ecode):
pathway = Pathway(self.graph)
pathway_curie = "{}:{}".format(pathway_prefix, pathway_id)
gene_curie = "{}:{}".format(component_prefix, component.strip())
eco_curie = eco_map[go_ecode]
pathway.addPathway(pathway_curie, pathway_label)
pathway.addComponentToPathway(gene_curie, pathway_curie)
association = Assoc(self.graph, self.name)
association.sub = gene_curie
association.rel = self.globaltt['involved in']
association.obj = pathway_curie
association.set_association_id()
association.add_evidence(eco_curie)
association.add_association_to_graph()
return
class CTD(Source):
"""
The Comparative Toxicogenomics Database (CTD) includes curated data
describing cross-species chemical–gene/protein interactions and
chemical– and gene–disease associations to illuminate molecular mechanisms
underlying variable susceptibility and environmentally influenced diseases.
Here, we fetch, parse, and convert data from CTD into triples,
leveraging only the associations based on DIRECT evidence
(not using the inferred associations).
We currently process the following associations:
* chemical-disease
* gene-pathway
* gene-disease
CTD curates relationships between genes and chemicals/diseases with
marker/mechanism and/or therapeutic.
Unfortunately, we cannot disambiguate between marker (gene expression) and
mechanism (causation) for these associations. Therefore, we are left to
relate these simply by "marker".
CTD also pulls in genes and pathway membership from KEGG and REACTOME.
We create groups of these following the pattern that the specific pathway
is a subclass of 'cellular process' (a go process), and the gene is
"involved in" that process.
For diseases, we preferentially use OMIM identifiers when they can be used
uniquely over MESH. Otherwise, we use MESH ids.
Note that we scrub the following identifiers and their associated data:
* REACT:REACT_116125 - generic disease class
* MESH:D004283 - dog diseases
* MESH:D004195 - disease models, animal
* MESH:D030342 - genetic diseases, inborn
* MESH:D040181 - genetic dieases, x-linked
* MESH:D020022 - genetic predisposition to a disease
"""
files = {
'chemical_disease_interactions': {
'file': 'CTD_chemicals_diseases.tsv.gz',
'url': 'http://ctdbase.org/reports/CTD_chemicals_diseases.tsv.gz'
},
'gene_pathway': {
'file': 'CTD_genes_pathways.tsv.gz',
'url': 'http://ctdbase.org/reports/CTD_genes_pathways.tsv.gz'
},
'gene_disease': {
'file': 'CTD_genes_diseases.tsv.gz',
'url': 'http://ctdbase.org/reports/CTD_genes_diseases.tsv.gz'
}
}
static_files = {
'publications': {'file': 'CTD_curated_references.tsv'}
}
def __init__(self, graph_type, are_bnodes_skolemized):
super().__init__(
graph_type,
are_bnodes_skolemized,
'ctd',
ingest_title='Comparative Toxicogenomics Database',
ingest_url='http://ctdbase.org',
license_url=None,
data_rights='http://ctdbase.org/about/legal.jsp'
# file_handle=None
)
if 'gene' not in self.all_test_ids:
LOG.warning("not configured with gene test ids.")
self.test_geneids = []
else:
self.test_geneids = self.all_test_ids['gene']
if 'disease' not in self.all_test_ids:
LOG.warning("not configured with disease test ids.")
self.test_diseaseids = []
else:
self.test_diseaseids = self.all_test_ids['disease']
self.geno = Genotype(self.graph)
self.pathway = Pathway(self.graph)
return
def fetch(self, is_dl_forced=False):
"""
Override Source.fetch()
Fetches resources from CTD using the CTD.files dictionary
Args:
:param is_dl_forced (bool): Force download
Returns:
:return None
"""
self.get_files(is_dl_forced)
self._fetch_disambiguating_assoc()
# consider creating subsets of the files that
# only have direct annotations (not inferred)
return
def parse(self, limit=None):
"""
Override Source.parse()
Parses version and interaction information from CTD
Args:
:param limit (int, optional) limit the number of rows processed
Returns:
:return None
"""
if limit is not None:
LOG.info("Only parsing first %d rows", limit)
LOG.info("Parsing files...")
# pub_map = dict()
# file_path = '/'.join((self.rawdir,
# self.static_files['publications']['file']))
# if os.path.exists(file_path) is True:
# pub_map = self._parse_publication_file(
# self.static_files['publications']['file']
# )
if self.test_only:
self.test_mode = True
self.geno = Genotype(self.graph)
self.pathway = Pathway(self.graph)
self._parse_ctd_file(
limit, self.files['chemical_disease_interactions']['file'])
self._parse_ctd_file(limit, self.files['gene_pathway']['file'])
self._parse_ctd_file(limit, self.files['gene_disease']['file'])
self._parse_curated_chem_disease(limit)
LOG.info("Done parsing files.")
return
def _parse_ctd_file(self, limit, file):
"""
Parses files in CTD.files dictionary
Args:
:param limit (int): limit the number of rows processed
:param file (str): file name (must be defined in CTD.file)
Returns:
:return None
"""
row_count = 0
version_pattern = re.compile(r'^# Report created: (.+)$')
is_versioned = False
file_path = '/'.join((self.rawdir, file))
with gzip.open(file_path, 'rt') as tsvfile:
reader = csv.reader(tsvfile, delimiter="\t")
for row in reader:
# Scan the header lines until we get the version
# There is no official version sp we are using
# the upload timestamp instead
if is_versioned is False:
match = re.match(version_pattern, ' '.join(row))
if match:
version = re.sub(r'\s|:', '-', match.group(1))
# TODO convert this timestamp to a proper timestamp
self.dataset.setVersion(version)
is_versioned = True
elif re.match(r'^#', ' '.join(row)):
pass
else:
row_count += 1
if file == self.files[
'chemical_disease_interactions']['file']:
self._process_interactions(row)
elif file == self.files['gene_pathway']['file']:
self._process_pathway(row)
elif file == self.files['gene_disease']['file']:
self._process_disease2gene(row)
if not self.test_mode and limit is not None and row_count >= limit:
break
return
def _process_pathway(self, row):
"""
Process row of CTD data from CTD_genes_pathways.tsv.gz
and generate triples
Args:
:param row (list): row of CTD data
Returns:
:return None
"""
model = Model(self.graph)
self._check_list_len(row, 4)
(gene_symbol, gene_id, pathway_name, pathway_id) = row
if self.test_mode and (int(gene_id) not in self.test_geneids):
return
entrez_id = 'NCBIGene:' + gene_id
pathways_to_scrub = [
'REACT:REACT_116125', # disease
"REACT:REACT_111045", # developmental biology
"REACT:REACT_200794", # Mus musculus biological processes
"REACT:REACT_13685"] # neuronal system ?
if pathway_id in pathways_to_scrub:
# these are lame "pathways" like generic
# "disease" and "developmental biology"
return
# convert KEGG pathway ids... KEGG:12345 --> KEGG-path:map12345
if re.match(r'KEGG', pathway_id):
pathway_id = re.sub(r'KEGG:', 'KEGG-path:map', pathway_id)
# just in case, add it as a class
model.addClassToGraph(entrez_id, None)
self.pathway.addPathway(pathway_id, pathway_name)
self.pathway.addGeneToPathway(entrez_id, pathway_id)
return
def _fetch_disambiguating_assoc(self):
"""
For any of the items in the chemical-disease association file that have
ambiguous association types we fetch the disambiguated associations
using the batch query API, and store these in a file. Elsewhere, we can
loop through the file and create the appropriate associations.
:return:
"""
disambig_file = '/'.join(
(self.rawdir, self.static_files['publications']['file']))
assoc_file = '/'.join(
(self.rawdir, self.files['chemical_disease_interactions']['file']))
# check if there is a local association file,
# and download if it's dated later than the original intxn file
if os.path.exists(disambig_file):
dfile_dt = os.stat(disambig_file)
afile_dt = os.stat(assoc_file)
if dfile_dt < afile_dt:
LOG.info(
"Local file date before chem-disease assoc file. "
" Downloading...")
else:
LOG.info(
"Local file date after chem-disease assoc file. "
" Skipping download.")
return
all_pubs = set()
dual_evidence = re.compile(r'^marker\/mechanism\|therapeutic$')
# first get all the unique publications
with gzip.open(assoc_file, 'rt') as tsvfile:
reader = csv.reader(tsvfile, delimiter="\t")
for row in reader:
if re.match(r'^#', ' '.join(row)):
continue
self._check_list_len(row, 10)
(chem_name, chem_id, cas_rn, disease_name, disease_id,
direct_evidence, inferred_gene_symbol, inference_score,
omim_ids, pubmed_ids) = row
if direct_evidence == '' or not \
re.match(dual_evidence, direct_evidence):
continue
if pubmed_ids is not None and pubmed_ids != '':
all_pubs.update(set(re.split(r'\|', pubmed_ids)))
sorted_pubs = sorted(list(all_pubs))
# now in batches of 4000, we fetch the chemical-disease associations
batch_size = 4000
params = {
'inputType': 'reference',
'report': 'diseases_curated',
'format': 'tsv',
'action': 'Download'
}
url = 'http://ctdbase.org/tools/batchQuery.go?q'
start = 0
end = min((batch_size, len(all_pubs))) # get them in batches of 4000
with open(disambig_file, 'wb') as dmbf:
while start < len(sorted_pubs):
params['inputTerms'] = '|'.join(sorted_pubs[start:end])
# fetch the data from url
LOG.info(
'fetching %d (%d-%d) refs: %s',
len(re.split(r'\|', params['inputTerms'])),
start, end, params['inputTerms'])
data = urllib.parse.urlencode(params)
encoding = 'utf-8'
binary_data = data.encode(encoding)
req = urllib.request.Request(url, binary_data)
resp = urllib.request.urlopen(req)
dmbf.write(resp.read())
start = end
end = min((start + batch_size, len(sorted_pubs)))
return
def _process_interactions(self, row):
"""
Process row of CTD data from CTD_chemicals_diseases.tsv.gz
and generate triples. Only create associations based on direct evidence
(not using the inferred-via-gene), and unambiguous relationships.
(Ambiguous ones will be processed in the sister method using the
disambiguated file). There are no OMIM ids for diseases in these cases,
so we associate with only the mesh disease ids.
Args:
:param row (list): row of CTD data
Returns:
:return None
"""
model = Model(self.graph)
self._check_list_len(row, 10)
(chem_name, chem_id, cas_rn, disease_name, disease_id, direct_evidence,
inferred_gene_symbol, inference_score, omim_ids, pubmed_ids) = row
if direct_evidence == '':
return
evidence_pattern = re.compile(r'^therapeutic|marker\/mechanism$')
# dual_evidence = re.compile(r'^marker\/mechanism\|therapeutic$')
# filter on those diseases that are mapped to omim ids in the test set
intersect = list(
set(['OMIM:' + str(i) for i in omim_ids.split('|')] +
[disease_id]) & set(self.test_diseaseids))
if self.test_mode and len(intersect) < 1:
return
chem_id = 'MESH:' + chem_id
reference_list = self._process_pubmed_ids(pubmed_ids)
if re.match(evidence_pattern, direct_evidence):
rel_id = self.resolve(direct_evidence)
model.addClassToGraph(chem_id, chem_name)
model.addClassToGraph(disease_id, None)
self._make_association(chem_id, disease_id, rel_id, reference_list)
else:
# there's dual evidence, but haven't mapped the pubs
pass
# LOG.debug(
# "Dual evidence for %s (%s) and %s (%s)",
# chem_name, chem_id, disease_name, disease_id)
return
def _process_disease2gene(self, row):
"""
Here, we process the disease-to-gene associations.
Note that we ONLY process direct associations
(not inferred through chemicals).
Furthermore, we also ONLY process "marker/mechanism" associations.
We preferentially utilize OMIM identifiers over MESH identifiers
for disease/phenotype.
Therefore, if a single OMIM id is listed under the "omim_ids" list,
we will choose this over any MeSH id that might be listed as
the disease_id. If multiple OMIM ids are listed in the omim_ids column,
we toss this for now.
(Mostly, we are not sure what to do with this information.)
We also pull in the MeSH labels here (but not OMIM) to ensure that
we have them (as they may not be brought in separately).
:param row:
:return:
"""
# if self.test_mode:
# graph = self.testgraph
# else:
# graph = self.graph
# self._check_list_len(row, 9)
# geno = Genotype(graph)
# gu = GraphUtils(curie_map.get())
model = Model(self.graph)
(gene_symbol, gene_id, disease_name, disease_id, direct_evidence,
inference_chemical_name, inference_score, omim_ids, pubmed_ids) = row
# we only want the direct associations; skipping inferred for now
if direct_evidence == '' or direct_evidence != 'marker/mechanism':
return
# scrub some of the associations...
# it seems odd to link human genes to the following "diseases"
diseases_to_scrub = [
'MESH:D004283', # dog diseases
'MESH:D004195', # disease models, animal
'MESH:D030342', # genetic diseases, inborn
'MESH:D040181', # genetic dieases, x-linked
'MESH:D020022'] # genetic predisposition to a disease
if disease_id in diseases_to_scrub:
LOG.info(
"Skipping association between NCBIGene:%s and %s",
str(gene_id), disease_id)
return
intersect = list(
set(['OMIM:' + str(i) for i in omim_ids.split('|')] +
[disease_id]) & set(self.test_diseaseids))
if self.test_mode and (
int(gene_id) not in self.test_geneids or len(intersect) < 1):
return
# there are three kinds of direct evidence:
# (marker/mechanism | marker/mechanism|therapeutic | therapeutic)
# we are only using the "marker/mechanism" for now
# TODO what does it mean for a gene to be therapeutic for disease?
# a therapeutic target?
gene_id = 'NCBIGene:' + gene_id
preferred_disease_id = disease_id
if omim_ids is not None and omim_ids != '':
omim_id_list = re.split(r'\|', omim_ids)
# If there is only one OMIM ID for the Disease ID
# or in the omim_ids list,
# use the OMIM ID preferentially over any MeSH ID.
if re.match(r'OMIM:.*', disease_id):
if len(omim_id_list) > 1:
# the disease ID is an OMIM ID and
# there is more than one OMIM entry in omim_ids.
# Currently no entries satisfy this condition
pass
elif disease_id != ('OMIM:' + omim_ids):
# the disease ID is an OMIM ID and
# there is only one non-equiv OMIM entry in omim_ids
# we preferentially use the disease_id here
LOG.warning(
"There may be alternate identifier for %s: %s",
disease_id, omim_ids)
# TODO: What should be done with the alternate disease IDs?
else:
if len(omim_id_list) == 1:
# the disease ID is not an OMIM ID
# and there is only one OMIM entry in omim_ids.
preferred_disease_id = 'OMIM:' + omim_ids
elif len(omim_id_list) > 1:
# This is when the disease ID is not an OMIM ID and
# there is more than one OMIM entry in omim_ids.
pass
model.addClassToGraph(gene_id, None)
# not sure if MESH is getting added separately.
# adding labels here for good measure
dlabel = None
if re.match(r'MESH', preferred_disease_id):
dlabel = disease_name
model.addClassToGraph(preferred_disease_id, dlabel)
# Add the disease to gene relationship.
rel_id = self.resolve(direct_evidence)
refs = self._process_pubmed_ids(pubmed_ids)
self._make_association(gene_id, preferred_disease_id, rel_id, refs)
return
def _make_association(self, subject_id, object_id, rel_id, pubmed_ids):
"""
Make a reified association given an array of pubmed identifiers.
Args:
:param subject_id id of the subject of the association (gene/chem)
:param object_id id of the object of the association (disease)
:param rel_id relationship id
:param pubmed_ids an array of pubmed identifiers
Returns:
:return None
"""
# TODO pass in the relevant Assoc class rather than relying on G2P
assoc = G2PAssoc(self.graph, self.name, subject_id, object_id, rel_id)
if pubmed_ids is not None and len(pubmed_ids) > 0:
for pmid in pubmed_ids:
ref = Reference(
self.graph, pmid, self.globaltt['journal article'])
ref.addRefToGraph()
assoc.add_source(pmid)
assoc.add_evidence(self.globaltt['traceable author statement'])
assoc.add_association_to_graph()
return
@staticmethod
def _process_pubmed_ids(pubmed_ids):
"""
Take a list of pubmed IDs and add PMID prefix
Args:
:param pubmed_ids - string representing publication
ids seperated by a | symbol
Returns:
:return list: Pubmed curies
"""
if pubmed_ids.strip() == '':
id_list = []
else:
id_list = pubmed_ids.split('|')
for (i, val) in enumerate(id_list):
id_list[i] = 'PMID:' + val
return id_list
def _parse_curated_chem_disease(self, limit):
model = Model(self.graph)
line_counter = 0
file_path = '/'.join(
(self.rawdir, self.static_files['publications']['file']))
with open(file_path, 'r') as tsvfile:
reader = csv.reader(tsvfile, delimiter="\t")
for row in reader:
# catch comment lines
if re.match(r'^#', ' '.join(row)):
continue
line_counter += 1
self._check_list_len(row, 10)
(pub_id, disease_label, disease_id, disease_cat, evidence,
chem_label, chem_id, cas_rn, gene_symbol, gene_acc) = row
if disease_id.strip() == '' or chem_id.strip() == '':
continue
rel_id = self.resolve(evidence)
chem_id = 'MESH:' + chem_id
model.addClassToGraph(chem_id, chem_label)
model.addClassToGraph(disease_id, None)
if pub_id != '':
pub_id = 'PMID:' + pub_id
ref = Reference(
self.graph, pub_id, ref_type=self.globaltt['journal article'])
ref.addRefToGraph()
pubids = [pub_id]
else:
pubids = None
self._make_association(chem_id, disease_id, rel_id, pubids)
if not self.test_mode and limit is not None and line_counter >= limit:
break
return
def getTestSuite(self):
import unittest
from tests.test_ctd import CTDTestCase
test_suite = unittest.TestLoader().loadTestsFromTestCase(CTDTestCase)
# test_suite.addTests(
# unittest.TestLoader().loadTestsFromTestCase(InteractionsTestCase))
return test_suite
class CTD(Source):
"""
The Comparative Toxicogenomics Database (CTD) includes curated data
describing cross-species chemical–gene/protein interactions and
chemical– and gene–disease associations to illuminate molecular mechanisms
underlying variable susceptibility and environmentally influenced diseases.
(updated monthly).
Here, we fetch, parse, and convert data from CTD into triples,
leveraging only the associations based on DIRECT evidence
(not using the inferred associations).
We currently process the following associations:
* chemical-disease
* gene-pathway
* gene-disease
CTD curates relationships between genes and chemicals/diseases with
'marker/mechanism' or 'therapeutic'. (observe strictly OR)
Unfortunately,
we cannot disambiguate between marker (gene expression) and
mechanism (causation) for these associations.
Therefore, we are left to relate these simply by "marker".
# We DISCONTIUED at some point prior to 202005
# CTD also pulls in genes and pathway membership from KEGG and REACTOME.
# We create groups of these following the pattern that the specific pathway
# is a subclass of 'cellular process' (a go process), and the gene is
# "involved in" that process.
For diseases, we preferentially use OMIM identifiers when they can be used
uniquely over MESH. Otherwise, we use MESH ids.
Note that we scrub the following identifiers and their associated data:
* REACT:REACT_116125 - generic disease class
* MESH:D004283 - dog diseases
* MESH:D004195 - disease models, animal
* MESH:D030342 - genetic diseases, inborn
* MESH:D040181 - genetic dieases, x-linked
* MESH:D020022 - genetic predisposition to a disease
"""
files = {
'chemical_disease_associations': {
'file':
'CTD_chemicals_diseases.tsv.gz',
'url':
'http://ctdbase.org/reports/CTD_chemicals_diseases.tsv.gz',
'columns': [ # expected
'ChemicalName',
'ChemicalID', # MeSH identifier
'CasRN', # CAS Registry Number, if available
'DiseaseName',
'DiseaseID', # MeSH or OMIM identifier
'DirectEvidence', # '|'-delimited list
'InferenceGeneSymbol',
'InferenceScore',
'OmimIDs', # '|'-delimited list
'PubMedIDs' # '|'-delimited list
]
},
# 'gene_pathway': {
# 'file': 'CTD_genes_pathways.tsv.gz',
# 'url': 'http://ctdbase.org/reports/CTD_genes_pathways.tsv.gz'
# },
# 'gene_disease': {
# 'file': 'CTD_genes_diseases.tsv.gz',
# 'url': 'https://ctdbase.org/reports/CTD_genes_diseases.tsv.gz'
# 'columns': [
# 'GeneSymbol',
# 'GeneID',
# 'DiseaseName',
# 'DiseaseID',
# 'DirectEvidence',
# 'InferenceChemicalName',
# 'InferenceScore',
# 'OmimIDs PubMedIDs'
# ]
# }
# 'chemicals': {
# 'file': 'CTD_chemicals.tsv.gz',
# 'url': 'https://ctdbase.org/reports/CTD_chemicals.tsv.gz',
# 'columns': [
# 'ChemicalName',
# 'ChemicalID', # Mesh
# 'CasRN',
# 'Definition',
# 'ParentIDs', # Mesh|... (note: cycles in graph ~250K edges)
# 'TreeNumbers',
# 'ParentTreeNumbers',
# 'Synonyms',
# 'DrugBankIDs'
# ]
# }
}
def __init__(self,
graph_type,
are_bnodes_skolemized,
data_release_version=None):
super().__init__(graph_type=graph_type,
are_bnodes_skized=are_bnodes_skolemized,
data_release_version=data_release_version,
name='ctd',
ingest_title='Comparative Toxicogenomics Database',
ingest_url='http://ctdbase.org',
ingest_logo='source-ctd.png',
license_url=None,
data_rights='http://ctdbase.org/about/legal.jsp'
# file_handle=None
)
if 'gene' not in self.all_test_ids:
LOG.warning("not configured with gene test ids.")
self.test_geneids = []
else:
self.test_geneids = self.all_test_ids['gene']
if 'disease' not in self.all_test_ids:
LOG.warning("not configured with disease test ids.")
self.test_diseaseids = []
else:
self.test_diseaseids = self.all_test_ids['disease']
self.geno = Genotype(self.graph)
self.pathway = Pathway(self.graph)
return
def fetch(self, is_dl_forced=False):
"""
Override Source.fetch()
Fetches resources from CTD using the CTD.files dictionary
Args:
:param is_dl_forced (bool): Force download
Returns:
:return None
"""
self.get_files(is_dl_forced)
src_key = 'chemical_disease_associations'
assoc_file = '/'.join((self.rawdir, self.files[src_key]['file']))
def parse(self, limit=None):
"""
Override Source.parse()
Parses version and interaction information from CTD
Args:
:param limit (int, optional) limit the number of rows processed
Returns:
:return None
"""
if limit is not None:
LOG.info("Only parsing first %d rows", limit)
LOG.info("Parsing files...")
if self.test_only:
self.test_mode = True
self.geno = Genotype(self.graph)
self.pathway = Pathway(self.graph)
src_key = 'chemical_disease_associations'
self._parse_ctd_file(limit, src_key)
# self._parse_ctd_file(limit, 'gene_pathway')
# self._parse_ctd_file(limit, 'gene_disease')
def _parse_ctd_file(self, limit, src_key):
"""
Parses files in CTD.files dictionary
Args:
:param limit (int): limit the number of rows processed
:param file (str): dictionary containing file name ('file') and ('url')
(must be defined in CTD.file)
Returns:
:return None
"""
row_count = 0
version_pattern = re.compile(r'^# Report created: (.+)$')
is_versioned = False
file_path = '/'.join((self.rawdir, self.files[src_key]['file']))
with gzip.open(file_path, 'rt') as tsvfile:
reader = csv.reader(tsvfile, delimiter="\t")
for row in reader:
# Scan the header lines until we get the version
# There is no official version sp we are using
# the upload timestamp instead
if is_versioned is False:
match = re.match(version_pattern, ' '.join(row))
if match:
version_date = re.sub(r'\s|:', '-', match.group(1))
self.dataset.set_ingest_source_file_version_date(
self.files[src_key]['url'], version_date)
is_versioned = True
elif re.match(r'^#', ' '.join(row)):
pass
else:
row_count += 1
if src_key == 'chemical_disease_associations':
self._process_interactions(row)
# elif file == self.files['gene_pathway']['file']:
# self._process_pathway(row)
# elif file == self.files['gene_disease']['file']:
# self._process_disease2gene(row)
if not self.test_mode and limit is not None and row_count >= limit:
break
return
def _process_pathway(self, row):
"""
Process row of CTD data from CTD_genes_pathways.tsv.gz
and generate triples
Args:
:param row (list): row of CTD data
Returns:
:return None
"""
model = Model(self.graph)
self._check_list_len(row, 4)
(gene_symbol, gene_id, pathway_name, pathway_id) = row
if self.test_mode and (int(gene_id) not in self.test_geneids):
return
entrez_id = 'NCBIGene:' + gene_id
pathways_to_scrub = [
'REACT:REACT_116125', # disease
"REACT:REACT_111045", # developmental biology
"REACT:REACT_200794", # Mus musculus biological processes
"REACT:REACT_13685"
] # neuronal system ?
if pathway_id in pathways_to_scrub:
# these are lame "pathways" like generic
# "disease" and "developmental biology"
return
# convert KEGG pathway ids... KEGG:12345 --> KEGG-path:map12345
if re.match(r'KEGG', pathway_id):
pathway_id = re.sub(r'KEGG:', 'KEGG-path:map', pathway_id)
# just in case, add it as a class
model.addType(entrez_id, self.globaltt['gene'])
self.pathway.addPathway(pathway_id, pathway_name)
self.pathway.addGeneToPathway(entrez_id, pathway_id)
return
def _process_interactions(self, row):
"""
Process row of CTD data from CTD_chemicals_diseases.tsv.gz
and generate triples. Only create associations based on direct evidence
(not using the inferred-via-gene), and unambiguous relationships.
(Ambiguous ones will be processed in the sister method using the
disambiguated file). There are no OMIM ids for diseases in these cases,
so we associate with only the mesh disease ids.
Args:
:param row (list): row of CTD data
Returns:
:return None
"""
model = Model(self.graph)
self._check_list_len(row, 10)
(chem_name, chem_id, cas_rn, disease_name, disease_id, direct_evidence,
inferred_gene_symbol, inference_score, omim_ids, pubmed_ids) = row
if direct_evidence == '':
return
evidence_pattern = re.compile(r'^therapeutic|marker\/mechanism$')
# filter on those diseases that are mapped to omim ids in the test set
intersect = list(
set(['OMIM:' + str(i) for i in omim_ids.split('|')] + [disease_id])
& set(self.test_diseaseids))
if self.test_mode and len(intersect) < 1:
return
chem_id = 'MESH:' + chem_id
reference_list = self._process_pubmed_ids(pubmed_ids)
rel_id = self.resolve(direct_evidence)
model.addClassToGraph(chem_id,
chem_name,
class_category=blv.terms['ChemicalSubstance'])
model.addClassToGraph(
disease_id,
None,
class_category=blv.terms['DiseaseOrPhenotypicFeature'])
self._make_association(chem_id, disease_id, rel_id, reference_list)
def _process_disease2gene(self, row):
"""
Here, we process the disease-to-gene associations.
Note that we ONLY process direct associations
(not inferred through chemicals).
Furthermore, we also ONLY process "marker/mechanism" associations.
We preferentially utilize OMIM identifiers over MESH identifiers
for disease/phenotype.
Therefore, if a single OMIM id is listed under the "omim_ids" list,
we will choose this over any MeSH id that might be listed as
the disease_id. If multiple OMIM ids are listed in the omim_ids column,
we toss this for now.
(Mostly, we are not sure what to do with this information.)
We also pull in the MeSH labels here (but not OMIM) to ensure that
we have them (as they may not be brought in separately).
:param row:
:return:
"""
# if self.test_mode:
# graph = self.testgraph
# else:
# graph = self.graph
# self._check_list_len(row, 9)
# geno = Genotype(graph)
# gu = GraphUtils(curie_map.get())
model = Model(self.graph)
(gene_symbol, gene_id, disease_name, disease_id, direct_evidence,
inference_chemical_name, inference_score, omim_ids, pubmed_ids) = row
# we only want the direct associations; skipping inferred for now
if direct_evidence == '' or direct_evidence != 'marker/mechanism':
return
# scrub some of the associations...
# it seems odd to link human genes to the following "diseases"
diseases_to_scrub = [
'MESH:D004283', # dog diseases
'MESH:D004195', # disease models, animal
'MESH:D030342', # genetic diseases, inborn
'MESH:D040181', # genetic dieases, x-linked
'MESH:D020022'
] # genetic predisposition to a disease
if disease_id in diseases_to_scrub:
LOG.info("Skipping association between NCBIGene:%s and %s",
str(gene_id), disease_id)
return
intersect = list(
set(['OMIM:' + str(i) for i in omim_ids.split('|')] + [disease_id])
& set(self.test_diseaseids))
if self.test_mode and (int(gene_id) not in self.test_geneids
or len(intersect) < 1):
return
# there are three kinds of direct evidence:
# (marker/mechanism | marker/mechanism|therapeutic | therapeutic)
# we are only using the "marker/mechanism" for now
# TODO what does it mean for a gene to be therapeutic for disease?
# a therapeutic target?
gene_id = 'NCBIGene:' + gene_id
preferred_disease_id = disease_id
if omim_ids is not None and omim_ids != '':
omim_id_list = re.split(r'\|', omim_ids)
# If there is only one OMIM ID for the Disease ID
# or in the omim_ids list,
# use the OMIM ID preferentially over any MeSH ID.
if re.match(r'OMIM:.*', disease_id):
if len(omim_id_list) > 1:
# the disease ID is an OMIM ID and
# there is more than one OMIM entry in omim_ids.
# Currently no entries satisfy this condition
pass
elif disease_id != ('OMIM:' + omim_ids):
# the disease ID is an OMIM ID and
# there is only one non-equiv OMIM entry in omim_ids
# we preferentially use the disease_id here
LOG.warning("There may be alternate identifier for %s: %s",
disease_id, omim_ids)
# TODO: What should be done with the alternate disease IDs?
else:
if len(omim_id_list) == 1:
# the disease ID is not an OMIM ID
# and there is only one OMIM entry in omim_ids.
preferred_disease_id = 'OMIM:' + omim_ids
elif len(omim_id_list) > 1:
# This is when the disease ID is not an OMIM ID and
# there is more than one OMIM entry in omim_ids.
pass
model.addClassToGraph(gene_id, None)
# not sure if MESH is getting added separately.
# adding labels here for good measure
dlabel = None
if re.match(r'MESH', preferred_disease_id):
dlabel = disease_name
model.addClassToGraph(preferred_disease_id, dlabel)
# Add the disease to gene relationship.
rel_id = self.resolve(direct_evidence)
refs = self._process_pubmed_ids(pubmed_ids)
self._make_association(gene_id, preferred_disease_id, rel_id, refs)
def _make_association(self,
subject_id,
object_id,
rel_id,
pubmed_ids,
subject_category=None,
object_category=None):
"""
Make a reified association given an array of pubmed identifiers.
Args:
:param subject_id id of the subject of the association (gene/chem)
:param object_id id of the object of the association (disease)
:param subject_category a biolink category CURIE for subject
:param object_category a biolink category CURIE for object
:param rel_id relationship id
:param pubmed_ids an array of pubmed identifiers
Returns:
:return None
"""
# TODO pass in the relevant Assoc class rather than relying on G2P
assoc = G2PAssoc(self.graph,
self.name,
subject_id,
object_id,
rel_id,
entity_category=subject_category,
phenotype_category=object_category)
if pubmed_ids is not None and len(pubmed_ids) > 0:
for pmid in pubmed_ids:
ref = Reference(self.graph, pmid,
self.globaltt['journal article'])
ref.addRefToGraph()
assoc.add_source(pmid)
assoc.add_evidence(self.globaltt['traceable author statement'])
assoc.add_association_to_graph()
@staticmethod
def _process_pubmed_ids(pubmed_ids):
"""
Take a list of pubmed IDs and add PMID prefix
Args:
:param pubmed_ids - string representing publication
ids seperated by a | symbol
Returns:
:return list: Pubmed curies
"""
if pubmed_ids.strip() == '':
id_list = []
else:
id_list = pubmed_ids.split('|')
for (i, val) in enumerate(id_list):
id_list[i] = 'PMID:' + val
return id_list
def getTestSuite(self):
import unittest
from tests.test_ctd import CTDTestCase
test_suite = unittest.TestLoader().loadTestsFromTestCase(CTDTestCase)
# test_suite.addTests(
# unittest.TestLoader().loadTestsFromTestCase(InteractionsTestCase))
return test_suite