biored.py

# coding=utf-8
# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""
Relation Extraction corpus with multiple entity types (e.g., gene/protein,
disease, chemical) and relation pairs (e.g., gene-disease; chemical-chemical),
on a set of 600 PubMed articles
"""

import itertools
import os
from collections import defaultdict
from typing import Dict, List, Tuple

import datasets
from bioc import pubtator

from .bigbiohub import kb_features
from .bigbiohub import BigBioConfig
from .bigbiohub import Tasks

_LANGUAGES = ["English"]
_PUBMED = True
_LOCAL = False
_CITATION = """\
@article{DBLP:journals/corr/abs-2204-04263,
  author    = {Ling Luo and
               Po{-}Ting Lai and
               Chih{-}Hsuan Wei and
               Cecilia N. Arighi and
               Zhiyong Lu},
  title     = {BioRED: {A} Comprehensive Biomedical Relation Extraction Dataset},
  journal   = {CoRR},
  volume    = {abs/2204.04263},
  year      = {2022},
  url       = {https://doi.org/10.48550/arXiv.2204.04263},
  doi       = {10.48550/arXiv.2204.04263},
  eprinttype = {arXiv},
  eprint    = {2204.04263},
  timestamp = {Wed, 11 May 2022 15:24:37 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2204-04263.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}
"""

_DATASETNAME = "biored"
_DISPLAYNAME = "BioRED"

_DESCRIPTION = """\
Relation Extraction corpus with multiple entity types (e.g., gene/protein,
disease, chemical) and relation pairs (e.g., gene-disease; chemical-chemical),
on a set of 600 PubMed articles
"""

_HOMEPAGE = "https://ftp.ncbi.nlm.nih.gov/pub/lu/BioRED/"

_LICENSE = "License information unavailable"

_URLS = {
    _DATASETNAME: "https://ftp.ncbi.nlm.nih.gov/pub/lu/BioRED/BIORED.zip",
}

_SUPPORTED_TASKS = [Tasks.NAMED_ENTITY_RECOGNITION, Tasks.NAMED_ENTITY_DISAMBIGUATION, Tasks.RELATION_EXTRACTION]

_SOURCE_VERSION = "1.0.0"

_BIGBIO_VERSION = "1.0.0"

logger = datasets.utils.logging.get_logger(__name__)


class BioredDataset(datasets.GeneratorBasedBuilder):
    """Relation Extraction corpus with multiple entity types (e.g., gene/protein, disease, chemical) and
    relation pairs (e.g., gene-disease; chemical-chemical), on a set of 600 PubMed articles"""

    # For bigbio_kb, this dataset uses a naming convention as
    # uid_[title/abstract/relation/entity_id]_[entity/relation_uid]

    SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
    BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)

    BUILDER_CONFIGS = [
        BigBioConfig(
            name=_DATASETNAME + "_source",
            version=SOURCE_VERSION,
            description=_DATASETNAME + " source schema",
            schema="source",
            subset_id=_DATASETNAME,
        ),
        BigBioConfig(
            name=_DATASETNAME + "_bigbio_kb",
            version=BIGBIO_VERSION,
            description=_DATASETNAME + " BigBio schema",
            schema="bigbio_kb",
            subset_id=_DATASETNAME,
        ),
    ]

    DEFAULT_CONFIG_NAME = _DATASETNAME + "_source"

    TYPE_TO_DATABASE = {
        "CellLine": "Cellosaurus",
        "ChemicalEntity": "MESH",
        "DiseaseOrPhenotypicFeature": "MESH",  # Some diseases are normalized to OMIM (~ handled by special rules)
        "GeneOrGeneProduct": "NCBIGene",
        "OrganismTaxon": "NCBITaxon",
        "SequenceVariant": "dbSNP",  # Not all variants are normalized to dbSNP (~ handled by special rules)
    }

    def _info(self) -> datasets.DatasetInfo:

        if self.config.schema == "source":

            features = datasets.Features(
                {
                    "pmid": datasets.Value("string"),
                    "passages": [
                        {
                            "type": datasets.Value("string"),
                            "text": datasets.Sequence(datasets.Value("string")),
                            "offsets": datasets.Sequence([datasets.Value("int32")]),
                        }
                    ],
                    "entities": [
                        {
                            "text": datasets.Sequence(datasets.Value("string")),
                            "offsets": datasets.Sequence([datasets.Value("int32")]),
                            "concept_id": datasets.Value("string"),
                            "semantic_type_id": datasets.Value("string"),
                        }
                    ],
                    "relations": [
                        {
                            "novel": datasets.Value("string"),
                            "type": datasets.Value("string"),
                            "concept_1": datasets.Value("string"),
                            "concept_2": datasets.Value("string"),
                        }
                    ],
                }
            )

        elif self.config.schema == "bigbio_kb":
            features = kb_features

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=str(_LICENSE),
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]:
        """Returns SplitGenerators."""

        urls = _URLS[_DATASETNAME]
        data_dir = dl_manager.download_and_extract(urls)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                # Whatever you put in gen_kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": os.path.join(data_dir, "BioRED", "Train.PubTator"),
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepath": os.path.join(data_dir, "BioRED", "Test.PubTator"),
                    "split": "test",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "filepath": os.path.join(data_dir, "BioRED", "Dev.PubTator"),
                    "split": "dev",
                },
            ),
        ]

    def _generate_examples(self, filepath, split: str) -> Tuple[int, Dict]:
        """Yields examples as (key, example) tuples."""

        if self.config.schema == "source":
            with open(filepath, "r", encoding="utf8") as fstream:
                for raw_document in self.generate_raw_docs(fstream):
                    document = self.parse_raw_doc(raw_document)
                    yield document["pmid"], document

        elif self.config.schema == "bigbio_kb":
            with open(filepath, "r", encoding="utf8") as fstream:
                uid = itertools.count(0)
                for raw_document in self.generate_raw_docs(fstream):
                    document = self.parse_raw_doc(raw_document)
                    pmid = str(document.pop("pmid"))
                    document["id"] = str(next(uid))
                    document["document_id"] = pmid

                    # Parse entities
                    entities = []
                    entity_id_to_mentions = defaultdict(list)  # Maps database ids to mention ids
                    for i, entity in enumerate(document["entities"]):
                        internal_id = pmid + "_" + str(i)

                        # Some entities are normalized to multiple database ids, therefore we
                        # may have multiple identifiers per mention
                        normalized_entity_ids = []
                        for database_id in entity["concept_id"].split(","):
                            database_id = database_id.strip()
                            entity_type = entity["semantic_type_id"]

                            # First check special db_name and database id assignment rules
                            if entity_type == "DiseaseOrPhenotypicFeature" and database_id.lower().startswith("omim"):
                                db_name = "OMIM"
                                database_id = database_id.split(":")[-1]
                            elif entity_type == "SequenceVariant" and not database_id.startswith("rs"):
                                db_name = "custom"

                            # If no special rule applies -> just take the default db_name for the entity type
                            else:
                                db_name = self.TYPE_TO_DATABASE[entity_type]

                            normalized_entity_ids.append({"db_name": db_name, "db_id": database_id})
                            entity_id_to_mentions[database_id].append(internal_id)

                        entities.append(
                            {
                                "id": internal_id,
                                "type": entity_type,
                                "text": entity["text"],
                                "normalized": normalized_entity_ids,
                                "offsets": entity["offsets"],
                            }
                        )

                    # BioRed provides abstract-level annotations for entity-linked relation pairs rather than
                    # materializing links between all surface form mentions of relation. For example document 11009181
                    # in train has (Positive_Correlation, D007980, D004409). Analogous to BC5CDR we enumerate all
                    # mention pairs concerning the entities in the triple.
                    relations = []
                    rel_uid = itertools.count(0)
                    for relation in document["relations"]:
                        head_mentions = entity_id_to_mentions[str(relation["concept_1"])]
                        tail_mentions = entity_id_to_mentions[str(relation["concept_2"])]

                        for head, tail in itertools.product(head_mentions, tail_mentions):
                            relations.append(
                                {
                                    "id": document["id"] + "_relation_" + str(next(rel_uid)),
                                    "type": relation["type"],
                                    "arg1_id": head,
                                    "arg2_id": tail,
                                    "normalized": [],
                                }
                            )

                    for passage in document["passages"]:
                        passage["id"] = document["id"] + "_" + passage["type"]

                    document["entities"] = entities
                    document["relations"] = relations
                    document["events"] = []
                    document["coreferences"] = []

                    yield document["document_id"], document

    def generate_raw_docs(self, fstream):
        """
        Given a filestream, this function yields documents from it
        """
        raw_document = []
        for line in fstream:
            if line.strip():
                raw_document.append(line.strip())
            elif raw_document:
                yield raw_document
                raw_document = []
        if raw_document:
            yield raw_document

    def parse_raw_doc(self, raw_doc):
        pmid, _, title = raw_doc[0].split("|")
        pmid = int(pmid)
        _, _, abstract = raw_doc[1].split("|")
        passages = [
            {"type": "title", "text": [title], "offsets": [[0, len(title)]]},
            {
                "type": "abstract",
                "text": [abstract],
                "offsets": [[len(title) + 1, len(title) + len(abstract) + 1]],
            },
        ]
        entities = []
        relations = []
        for line in raw_doc[2:]:
            mentions = line.split("\t")
            (_pmid, _type_ind, *rest) = mentions
            if _type_ind in [
                "Positive_Correlation",
                "Association",
                "Negative_Correlation",
                "Bind",
                "Conversion",
                "Cotreatment",
                "Cause",
                "Comparison",
                "Drug_Interaction",
            ]:
                # Relations handled here
                relation_type = _type_ind
                concept_1, concept_2, novel = rest
                relation = {
                    "type": relation_type,
                    "concept_1": concept_1,
                    "concept_2": concept_2,
                    "novel": novel,
                }
                relations.append(relation)
            elif _type_ind.isnumeric():
                # Entities handled here
                start_idx = _type_ind
                end_idx, mention, semantic_type_id, entity_ids = rest
                entities.append(
                    {
                        "offsets": [[int(start_idx), int(end_idx)]],
                        "text": [mention],
                        "semantic_type_id": semantic_type_id,
                        "concept_id": entity_ids,
                    }
                )
            else:
                logger.warn(f"Skipping annotation in Document ID: {_pmid}. Unexpected format")
        return {
            "pmid": pmid,
            "passages": passages,
            "entities": entities,
            "relations": relations,
        }