### ATTENTION!!!! PLEASE PIPE THE OUTPUT OF THIS SCRIPT THROUGH sort | uniq !!! ###
### Doing it within python is a waste of resources. Linux does it much faster. ###
def get_parents(bottom_id, dag, root_id='HP:0000118'):
if bottom_id == root_id:
return set([bottom_id])
rv = set()
if bottom_id in dag.edges:
for parent in dag.edges[bottom_id]:
rv |= get_parents(parent, dag)
rv.add(bottom_id)
return rv
if __name__ == '__main__':
hpo_dag = dutil.read_hpo_dag()
with open('%s/onto/data/hpo_phenotypes.tsv' % APP_HOME) as f:
for line in f:
toks = line.strip().split('\t')
hpo_id = toks[0]
pheno_name = toks[1]
parent_ids = get_parents(hpo_id,
hpo_dag) # includes the original hpo_id
assert hpo_id in parent_ids
if 'HP:0000118' not in parent_ids:
continue
sys.stdout.write(hpo_id + '\t' + pheno_name + '\n')
sys.stdout.flush()
def main(id_file, candidate_file):
# Load list of all pubmed IDs in the dataset
print >> sys.stderr, 'Loading list of pubmed IDs from doc ID list.'
doi_to_pmid = dutil.read_doi_to_pmid()
pmids_in_data = set()
num_docs = 0
with open(id_file) as f:
for line in f:
doc_id = line.strip()
pmid = dutil.get_pubmed_id_for_doc(doc_id, doi_to_pmid=doi_to_pmid)
if pmid:
pmids_in_data.add(pmid)
num_docs += 1
print >> sys.stderr, '%d/%d documents have PubMed IDs.' % (
len(pmids_in_data), num_docs)
# Load map from Pubmed ID to HPO term via MeSH
print >> sys.stderr, 'Loading supervision data via MeSH'
mesh_supervision = collections.defaultdict(set)
with open('%s/onto/data/hpo_to_pmid_via_mesh.tsv' % util.APP_HOME) as f:
for line in f:
hpo_id, pmid = line.strip().split('\t')
if pmid in pmids_in_data:
mesh_supervision[pmid].add(hpo_id)
# Identify all true pairs from MeSH
true_pairs = set()
for pmid in pmids_in_data:
for hpo in mesh_supervision[pmid]:
true_pairs.add((pmid, hpo))
# Load map from Pubmed ID to HPO term based on extracted candidates
print >> sys.stderr, 'Loading extracted pheno candidates'
candidates = collections.defaultdict(set)
with open(candidate_file) as f:
for line in f:
doc_id, hpo_id = line.strip().split('\t')
pmid = dutil.get_pubmed_id_for_doc(doc_id, doi_to_pmid=doi_to_pmid)
if pmid:
candidates[pmid].add(hpo_id)
# Load HPO DAG
# We say we found a HPO term if we find either the exact HPO term
# or one of its children
hpo_dag = dutil.read_hpo_dag()
# Determine which true pairs had candidate mentions for them
found_pairs = set()
missed_pairs = set()
for pmid, hpo in true_pairs:
found_hpo_ids = candidates[pmid]
for cand_hpo in found_hpo_ids:
if cand_hpo == '\N': continue
if hpo_dag.has_child(hpo, cand_hpo):
found_pairs.add((pmid, hpo))
break
else:
missed_pairs.add((pmid, hpo))
# Compute recall
num_true = len(true_pairs)
num_found = len(found_pairs)
print >> sys.stderr, 'Recall: %d/%d = %g' % (
num_found, num_true, float(num_found) / num_true)
# Compute other statistics
num_article = len(pmids_in_data)
num_annotated = sum(1 for x in pmids_in_data if len(mesh_supervision[x]) > 0)
print >> sys.stderr, '%d/%d = %g pubmed articles had HPO annotation' % (
num_annotated, num_article, float(num_annotated) / num_article)
# Read in HPO information
hpo_info_dict = dict()
with open('%s/onto/data/hpo_phenotypes.tsv' % util.APP_HOME) as f:
for line in f:
toks = line.strip('\r\n').split('\t')
hpo_id = toks[0]
hpo_info_dict[hpo_id] = toks[0:3]
# Sample some error cases
missed_sample = random.sample(list(missed_pairs), 100)
for pmid, hpo in missed_sample:
hpo_info = hpo_info_dict[hpo]
pubmed_url = 'http://www.ncbi.nlm.nih.gov/pubmed/%s' % pmid
hpo_url = 'www.human-phenotype-ontology.org/hpoweb/showterm?id=%s' % hpo
toks = [pubmed_url, hpo_url] + hpo_info
print '\t'.join(toks)
('section_id', 'text'), ('sent_id', 'int'),
('gene_mention_id', 'text'), ('gene_name', 'text'),
('gene_wordidxs', 'int[]'),
('gene_is_correct', 'boolean'),
('pheno_mention_id', 'text'),
('pheno_entity', 'text'), ('pheno_wordidxs', 'int[]'),
('pheno_is_correct', 'boolean'), ('words', 'text[]'),
('lemmas', 'text[]'), ('poses', 'text[]'),
('ners', 'text[]'), ('dep_paths', 'text[]'),
('dep_parents', 'int[]')])
# This defines the output Relation object
Feature = collections.namedtuple(
'Feature', ['doc_id', 'section_id', 'relation_id', 'name'])
HPO_DAG = dutil.read_hpo_dag()
def replace_opts(opts, replaceList):
ret = {}
for name in opts:
strings = opts[name]
for (pattern, subst) in replaceList:
if name.endswith('rgx'):
subst = re.escape(subst)
strings = [s.replace(pattern, subst) for s in strings]
ret[name] = strings
return ret
CACHE = {}
#! /usr/bin/env python
from data_util import get_hpo_phenos, get_parents, read_hpo_dag, read_hpo_synonyms
if __name__ == "__main__":
hpo_dag = read_hpo_dag()
names = read_hpo_synonyms(1)
synonyms = read_hpo_synonyms()
allowed_phenos = set(get_hpo_phenos(hpo_dag))
for hpo_id in allowed_phenos.copy():
parent_ids = get_parents(hpo_id, hpo_dag) # includes the original hpo_id
assert hpo_id in parent_ids
if 'HP:0000118' not in parent_ids:
sys.stderr.write('"{0}": not a phenotypic abnormality\n'.format(hpo_id.strip()))
continue
parent_ids.remove('HP:0000118')
for parent_id in parent_ids:
allowed_phenos.add(parent_id)
for hpo_id in allowed_phenos:
print "%s\t%s\t%s" % (hpo_id, '|^|'.join(names[hpo_id]), '|^|'.join(synonyms[hpo_id]))
