def _preprocess(
examples: List[str],
sort_rels: bool = True,
entity_hinting: Optional[EntityHinting] = None,
) -> List[str]:
kwargs = {
"concepts": ["gene", "disease"],
"skip_malformed": True
} if entity_hinting else {}
abstracts, anns, labels = examples
pubtator_content = _convert_to_pubtator(abstracts=abstracts,
anns=anns,
labels=labels)
pubtator_annotations = util.parse_pubtator(
pubtator_content=pubtator_content,
text_segment=util.TextSegment.both,
)
seq2rel_annotations = util.pubtator_to_seq2rel(
pubtator_annotations,
sort_rels=sort_rels,
entity_hinting=entity_hinting,
**kwargs)
return seq2rel_annotations
def _preprocess(
pubtator_content: str,
sort_rels: bool = True,
entity_hinting: Optional[EntityHinting] = None,
filter_hypernyms: bool = False,
) -> List[str]:
kwargs = {
"concepts": ["chemical", "disease"],
"skip_malformed": True
} if entity_hinting else {}
pubtator_annotations = util.parse_pubtator(
pubtator_content=pubtator_content,
text_segment=util.TextSegment.both,
)
# This is unique the the CDR corpus, which contains many negative relations that are
# actually valid, but are not annotated because they contain a disease entity which is the
# hypernym of a disease entity in a positive relation. We need to filter these out before
# evaluation, so this function finds all such cases and adds them to the filtered_relations
# field of the annoations. See: https://arxiv.org/abs/1909.00228 for details.
if filter_hypernyms:
_filter_hypernyms(pubtator_annotations)
seq2rel_annotations = util.pubtator_to_seq2rel(
pubtator_annotations,
sort_rels=sort_rels,
entity_hinting=entity_hinting,
**kwargs,
)
return seq2rel_annotations
def test_parse_pubtator_compound_ent() -> None:
# A truncated example taken from the CDR dataset
pmid = "17854040"
title_text = (
"Mutations associated with lamivudine-resistance in therapy-na ve hepatitis B virus (HBV)"
" infected patients with and without HIV co-infection: implications for antiretroviral"
" therapy in HBV and HIV co-infected South African patients. infected patients with and"
" without HIV co-infection: implications for antiretroviral therapy in HBV and HIV"
" co-infected South African patients.")
abstract_text = (
"This was an exploratory study to investigate lamivudine-resistant hepatitis B virus (HBV)"
" strains in selected lamivudine-na ve HBV carriers with and without human"
" immunodeficiency virus (HIV) co-infection in South African patients. Thirty-five"
" lamivudine-naive HBV infected patients with or without HIV co-infection were studied: 15"
" chronic HBV mono-infected patients and 20 HBV-HIV co-infected patients."
)
pubtator_content = f"""
{pmid}|t|{title_text}
{pmid}|a|{abstract_text}
{pmid}\t26\t36\tlamivudine\tChemical\tD019259
{pmid}\t59\t61\tna\tChemical\tD012964
{pmid}\t66\t98\thepatitis B virus (HBV) infected\tDisease\tD006509
{pmid}\t125\t141\tHIV co-infection\tDisease\tD015658
{pmid}\t186\t209\tHBV and HIV co-infected\tDisease\tD006509|D015658 HBV infected|HIV infected
"""
expected = schemas.PubtatorAnnotation(
pmid=pmid,
text=f"{title_text} {abstract_text}",
entities={
"D019259":
schemas.PubtatorEntity(
mentions=["lamivudine"],
offsets=[(26, 36)],
label="Chemical",
),
"D012964":
schemas.PubtatorEntity(mentions=["na"],
offsets=[(59, 61)],
label="Chemical"),
"D006509":
schemas.PubtatorEntity(
mentions=["hepatitis B virus (HBV) infected", "HBV infected"],
offsets=[(66, 98), (186, 209)],
label="Disease",
),
"D015658":
schemas.PubtatorEntity(
mentions=["HIV co-infection", "HIV infected"],
offsets=[(125, 141), (194, 209)],
label="Disease",
),
},
)
actual = util.parse_pubtator(pubtator_content,
text_segment=util.TextSegment.both)
assert actual[0].text == expected.text
assert actual[0].entities == expected.entities
assert actual[0].relations == expected.relations
def test_parse_pubtator_no_abstract_raises_value_error() -> None:
# A truncated example taken from the CDR dataset
pmid = "2339463"
title_text = "Cerebral sinus thrombosis as a potential hazard of antifibrinolytic treatment in menorrhagia."
abstract_text = ""
pubtator_content = f"""
{pmid}|t|{title_text}
{pmid}|a|{abstract_text}
{pmid}\t0\t25\tCerebral sinus thrombosis\tDisease
"""
# A ValueError should be raised if no abstract provided but text_segment is "abstract" or "both"
with pytest.raises(ValueError):
_ = util.parse_pubtator(pubtator_content,
text_segment=util.TextSegment.abstract)
with pytest.raises(ValueError):
_ = util.parse_pubtator(pubtator_content,
text_segment=util.TextSegment.both)
def test_parse_pubtator_skip_malformed_raises_value_error() -> None:
# A truncated example taken from the CDR dataset
pmid = "2339463"
title_text = "Cerebral sinus thrombosis as a potential hazard of antifibrinolytic treatment in menorrhagia."
abstract_text = (
"We describe a 42-year-old woman who developed superior sagittal and left transverse sinus"
" thrombosis associated with prolonged epsilon-aminocaproic acid therapy for menorrhagia."
)
pubtator_content = f"""
{pmid}|t|{title_text}
{pmid}|a|{abstract_text}
{pmid}\t0\t25\tCerebral sinus thrombosis\tDisease
"""
# There should be no error when skip_malformed is True...
_ = util.parse_pubtator(pubtator_content, skip_malformed=True)
# ...and and error when it is False
with pytest.raises(ValueError):
_ = util.parse_pubtator(pubtator_content, skip_malformed=False)
def _preprocess(
examples: List[Dict[str, Any]],
rel_labels: Optional[Dict[str, str]] = None,
sort_rels: bool = True,
) -> List[str]:
pubtator_content = _convert_to_pubtator(examples, rel_labels=rel_labels)
pubtator_annotations = util.parse_pubtator(
pubtator_content=pubtator_content,
text_segment=util.TextSegment.abstract,
)
seq2rel_annotations = util.pubtator_to_seq2rel(pubtator_annotations,
sort_rels=sort_rels)
return seq2rel_annotations
def _preprocess(
examples: List[Dict[str, Any]],
sort_rels: bool = True,
entity_hinting: Optional[EntityHinting] = None,
) -> List[str]:
kwargs = ({
"concepts": ["chemical", "gene", "mutation"],
"skip_malformed": True
} if entity_hinting else {})
pubtator_content = _convert_to_pubtator(examples)
pubtator_annotations = util.parse_pubtator(
pubtator_content=pubtator_content,
text_segment=util.TextSegment.abstract,
)
seq2rel_annotations = util.pubtator_to_seq2rel(
pubtator_annotations,
sort_rels=sort_rels,
entity_hinting=entity_hinting,
**kwargs)
return seq2rel_annotations
def test_parse_pubtator() -> None:
# A truncated example taken from the CDR dataset
pmid = "18020536"
title_text = (
"Associations between use of benzodiazepines or related drugs and health, physical"
" abilities and cognitive function: a non-randomised clinical study in the elderly."
)
abstract_text = (
"OBJECTIVE: To describe associations between the use of benzodiazepines or related drugs"
" (BZDs/RDs) and health, functional abilities and cognitive function in the elderly."
" METHODS: A non-randomised clinical study of patients aged > or =65 years admitted to"
" acute hospital wards during 1 month. 164 patients (mean age +/- standard deviation [SD]"
" 81.6 +/- 6.8 years) were admitted. Of these, nearly half (n = 78) had used BZDs/RDs"
" before admission, and the remainder (n = 86) were non-users. Cognitive ability was"
" assessed by the Mini-Mental State Examination (MMSE). Patients scoring > or =20 MMSE"
" sum points were interviewed (n = 79) and questioned regarding symptoms and functional"
" abilities during the week prior to admission.")
# Include a dummy annotation with ID == -1. These should be ignored.
pubtator_content = f"""
{pmid}|t|{title_text}
{pmid}|a|{abstract_text}
{pmid}\t28\t43\tbenzodiazepines\tChemical\tD001569
{pmid}\t219\t234\tbenzodiazepines\tChemical\tD001569
{pmid}\t253\t257\tBZDs\tChemical\tD001569
{pmid}\t583\t587\tBZDs\tChemical\tD001569
{pmid}\t1817\t1826\ttiredness\tDisease\tD005221
{pmid}\t0\t0\tArbitrary\tArbitrary\t-1
{pmid}\tCID\tD001569\tD005221
"""
title_entities = {
"D001569":
schemas.PubtatorEntity(
mentions=["benzodiazepines"],
offsets=[(28, 43)],
label="Chemical",
),
}
abstract_entities = {
"D001569":
schemas.PubtatorEntity(
mentions=["benzodiazepines", "BZDs", "BZDs"],
offsets=[(219, 234), (253, 257), (583, 587)],
label="Chemical",
),
"D005221":
schemas.PubtatorEntity(mentions=["tiredness"],
offsets=[(1817, 1826)],
label="Disease"),
}
both_entities = {
"D001569":
schemas.PubtatorEntity(
mentions=title_entities["D001569"].mentions +
abstract_entities["D001569"].mentions,
offsets=title_entities["D001569"].offsets +
abstract_entities["D001569"].offsets,
label="Chemical",
),
"D005221":
schemas.PubtatorEntity(mentions=["tiredness"],
offsets=[(1817, 1826)],
label="Disease"),
}
# Title only
expected = schemas.PubtatorAnnotation(pmid=pmid,
text=title_text,
entities=title_entities,
relations=[])
actual = util.parse_pubtator(pubtator_content,
text_segment=util.TextSegment.title)
# Breaking up the asserts leads to much clearer outputs when the test fails
assert actual[0].text == expected.text
assert actual[0].entities == expected.entities
assert actual[0].relations == expected.relations
# Abstract only
expected = schemas.PubtatorAnnotation(
pmid=pmid,
text=abstract_text,
entities=abstract_entities,
relations=[("D001569", "D005221", "CID")],
)
actual = util.parse_pubtator(pubtator_content,
text_segment=util.TextSegment.abstract)
assert actual[0].text == expected.text
assert actual[0].entities == expected.entities
assert actual[0].relations == expected.relations
# Both
expected = schemas.PubtatorAnnotation(
pmid=pmid,
text=f"{title_text} {abstract_text}",
entities=both_entities,
relations=[("D001569", "D005221", "CID")],
)
actual = util.parse_pubtator(pubtator_content,
text_segment=util.TextSegment.both)
assert actual[0].text == expected.text
assert actual[0].entities == expected.entities
assert actual[0].relations == expected.relations