gabrielaltay commited on
Commit
6dae873
1 Parent(s): 27d596a

upload hubscripts/chemdner_hub.py to hub from bigbio repo

Browse files
Files changed (1) hide show
  1. chemdner.py +417 -0
chemdner.py ADDED
@@ -0,0 +1,417 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ import os
16
+ import re
17
+ from typing import Dict, Iterator, List, Tuple
18
+
19
+ import bioc
20
+ import datasets
21
+ from bioc import biocxml
22
+
23
+ from .bigbiohub import kb_features
24
+ from .bigbiohub import BigBioConfig
25
+ from .bigbiohub import Tasks
26
+
27
+ _LANGUAGES = ['English']
28
+ _PUBMED = True
29
+ _LOCAL = False
30
+ _CITATION = """\
31
+ @article{Krallinger2015,
32
+ title = {The CHEMDNER corpus of chemicals and drugs and its annotation principles},
33
+ author = {
34
+ Krallinger, Martin and Rabal, Obdulia and Leitner, Florian and Vazquez,
35
+ Miguel and Salgado, David and Lu, Zhiyong and Leaman, Robert and Lu, Yanan
36
+ and Ji, Donghong and Lowe, Daniel M. and Sayle, Roger A. and
37
+ Batista-Navarro, Riza Theresa and Rak, Rafal and Huber, Torsten and
38
+ Rockt{\"a}schel, Tim and Matos, S{\'e}rgio and Campos, David and Tang,
39
+ Buzhou and Xu, Hua and Munkhdalai, Tsendsuren and Ryu, Keun Ho and Ramanan,
40
+ S. V. and Nathan, Senthil and {\v{Z}}itnik, Slavko and Bajec, Marko and
41
+ Weber, Lutz and Irmer, Matthias and Akhondi, Saber A. and Kors, Jan A. and
42
+ Xu, Shuo and An, Xin and Sikdar, Utpal Kumar and Ekbal, Asif and Yoshioka,
43
+ Masaharu and Dieb, Thaer M. and Choi, Miji and Verspoor, Karin and Khabsa,
44
+ Madian and Giles, C. Lee and Liu, Hongfang and Ravikumar, Komandur
45
+ Elayavilli and Lamurias, Andre and Couto, Francisco M. and Dai, Hong-Jie
46
+ and Tsai, Richard Tzong-Han and Ata, Caglar and Can, Tolga and Usi{\'e},
47
+ Anabel and Alves, Rui and Segura-Bedmar, Isabel and Mart{\'i}nez, Paloma
48
+ and Oyarzabal, Julen and Valencia, Alfonso
49
+ },
50
+ year = 2015,
51
+ month = {Jan},
52
+ day = 19,
53
+ journal = {Journal of Cheminformatics},
54
+ volume = 7,
55
+ number = 1,
56
+ pages = {S2},
57
+ doi = {10.1186/1758-2946-7-S1-S2},
58
+ issn = {1758-2946},
59
+ url = {https://doi.org/10.1186/1758-2946-7-S1-S2},
60
+ abstract = {
61
+ The automatic extraction of chemical information from text requires the
62
+ recognition of chemical entity mentions as one of its key steps. When
63
+ developing supervised named entity recognition (NER) systems, the
64
+ availability of a large, manually annotated text corpus is desirable.
65
+ Furthermore, large corpora permit the robust evaluation and comparison of
66
+ different approaches that detect chemicals in documents. We present the
67
+ CHEMDNER corpus, a collection of 10,000 PubMed abstracts that contain a
68
+ total of 84,355 chemical entity mentions labeled manually by expert
69
+ chemistry literature curators, following annotation guidelines specifically
70
+ defined for this task. The abstracts of the CHEMDNER corpus were selected
71
+ to be representative for all major chemical disciplines. Each of the
72
+ chemical entity mentions was manually labeled according to its
73
+ structure-associated chemical entity mention (SACEM) class: abbreviation,
74
+ family, formula, identifier, multiple, systematic and trivial. The
75
+ difficulty and consistency of tagging chemicals in text was measured using
76
+ an agreement study between annotators, obtaining a percentage agreement of
77
+ 91. For a subset of the CHEMDNER corpus (the test set of 3,000 abstracts)
78
+ we provide not only the Gold Standard manual annotations, but also mentions
79
+ automatically detected by the 26 teams that participated in the BioCreative
80
+ IV CHEMDNER chemical mention recognition task. In addition, we release the
81
+ CHEMDNER silver standard corpus of automatically extracted mentions from
82
+ 17,000 randomly selected PubMed abstracts. A version of the CHEMDNER corpus
83
+ in the BioC format has been generated as well. We propose a standard for
84
+ required minimum information about entity annotations for the construction
85
+ of domain specific corpora on chemical and drug entities. The CHEMDNER
86
+ corpus and annotation guidelines are available at:
87
+ ttp://www.biocreative.org/resources/biocreative-iv/chemdner-corpus/
88
+ }
89
+ }
90
+ """
91
+
92
+ _DESCRIPTION = """\
93
+ We present the CHEMDNER corpus, a collection of 10,000 PubMed abstracts that
94
+ contain a total of 84,355 chemical entity mentions labeled manually by expert
95
+ chemistry literature curators, following annotation guidelines specifically
96
+ defined for this task. The abstracts of the CHEMDNER corpus were selected to be
97
+ representative for all major chemical disciplines. Each of the chemical entity
98
+ mentions was manually labeled according to its structure-associated chemical
99
+ entity mention (SACEM) class: abbreviation, family, formula, identifier,
100
+ multiple, systematic and trivial.
101
+ """
102
+
103
+ _DATASETNAME = "chemdner"
104
+ _DISPLAYNAME = "CHEMDNER"
105
+
106
+ _HOMEPAGE = "https://biocreative.bioinformatics.udel.edu/resources/biocreative-iv/chemdner-corpus/"
107
+
108
+ _LICENSE = 'License information unavailable'
109
+
110
+ _URLs = {
111
+ "source": "https://ftp.ncbi.nlm.nih.gov/pub/lu/BC7-NLM-Chem-track/BC7T2-CHEMDNER-corpus_v2.BioC.xml.gz",
112
+ "bigbio_kb": "https://ftp.ncbi.nlm.nih.gov/pub/lu/BC7-NLM-Chem-track/BC7T2-CHEMDNER-corpus_v2.BioC.xml.gz",
113
+ "bigbio_text": "https://ftp.ncbi.nlm.nih.gov/pub/lu/BC7-NLM-Chem-track/BC7T2-CHEMDNER-corpus_v2.BioC.xml.gz",
114
+ }
115
+
116
+ _SUPPORTED_TASKS = [
117
+ Tasks.NAMED_ENTITY_RECOGNITION,
118
+ Tasks.TEXT_CLASSIFICATION,
119
+ ]
120
+ _SOURCE_VERSION = "1.0.0"
121
+ _BIGBIO_VERSION = "1.0.0"
122
+
123
+
124
+ class CHEMDNERDataset(datasets.GeneratorBasedBuilder):
125
+ """CHEMDNER"""
126
+
127
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
128
+ BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
129
+
130
+ BUILDER_CONFIGS = [
131
+ BigBioConfig(
132
+ name="chemdner_source",
133
+ version=SOURCE_VERSION,
134
+ description="CHEMDNER source schema",
135
+ schema="source",
136
+ subset_id="chemdner",
137
+ ),
138
+ BigBioConfig(
139
+ name="chemdner_bigbio_kb",
140
+ version=BIGBIO_VERSION,
141
+ description="CHEMDNER BigBio schema (KB)",
142
+ schema="bigbio_kb",
143
+ subset_id="chemdner",
144
+ ),
145
+ BigBioConfig(
146
+ name="chemdner_bigbio_text",
147
+ version=BIGBIO_VERSION,
148
+ description="CHEMDNER BigBio schema (TEXT)",
149
+ schema="bigbio_text",
150
+ subset_id="chemdner",
151
+ ),
152
+ ]
153
+
154
+ DEFAULT_CONFIG_NAME = "chemdner_source"
155
+
156
+ def _info(self):
157
+
158
+ if self.config.schema == "source":
159
+ # this is a variation on the BioC format
160
+ features = datasets.Features(
161
+ {
162
+ "passages": [
163
+ {
164
+ "document_id": datasets.Value("string"),
165
+ "type": datasets.Value("string"),
166
+ "text": datasets.Value("string"),
167
+ "offset": datasets.Value("int32"),
168
+ "entities": [
169
+ {
170
+ "id": datasets.Value("string"),
171
+ "offsets": [[datasets.Value("int32")]],
172
+ "text": [datasets.Value("string")],
173
+ "type": datasets.Value("string"),
174
+ "normalized": [
175
+ {
176
+ "db_name": datasets.Value("string"),
177
+ "db_id": datasets.Value("string"),
178
+ }
179
+ ],
180
+ }
181
+ ],
182
+ }
183
+ ]
184
+ }
185
+ )
186
+
187
+ elif self.config.schema == "bigbio_kb":
188
+ features = kb_features
189
+
190
+ elif self.config.schema == "bigbio_text":
191
+ features = text_features
192
+
193
+ return datasets.DatasetInfo(
194
+ description=_DESCRIPTION,
195
+ features=features,
196
+ supervised_keys=None,
197
+ homepage=_HOMEPAGE,
198
+ license=str(_LICENSE),
199
+ citation=_CITATION,
200
+ )
201
+
202
+ def _split_generators(self, dl_manager):
203
+ """Returns SplitGenerators."""
204
+
205
+ my_urls = _URLs[self.config.schema]
206
+ data_dir = dl_manager.download_and_extract(my_urls) + "/"
207
+ return [
208
+ datasets.SplitGenerator(
209
+ name=datasets.Split.TRAIN,
210
+ # These kwargs will be passed to _generate_examples
211
+ gen_kwargs={
212
+ "filepath": os.path.join(
213
+ data_dir, "BC7T2-CHEMDNER-corpus-training.BioC.xml"
214
+ ),
215
+ "split": "train",
216
+ },
217
+ ),
218
+ datasets.SplitGenerator(
219
+ name=datasets.Split.TEST,
220
+ # These kwargs will be passed to _generate_examples
221
+ gen_kwargs={
222
+ "filepath": os.path.join(
223
+ data_dir, "BC7T2-CHEMDNER-corpus-evaluation.BioC.xml"
224
+ ),
225
+ "split": "test",
226
+ },
227
+ ),
228
+ datasets.SplitGenerator(
229
+ name=datasets.Split.VALIDATION,
230
+ # These kwargs will be passed to _generate_examples
231
+ gen_kwargs={
232
+ "filepath": os.path.join(
233
+ data_dir, "BC7T2-CHEMDNER-corpus-development.BioC.xml"
234
+ ),
235
+ "split": "dev",
236
+ },
237
+ ),
238
+ ]
239
+
240
+ def _get_passages_and_entities(
241
+ self, d: bioc.BioCDocument
242
+ ) -> Tuple[List[Dict], List[List[Dict]]]:
243
+
244
+ passages: List[Dict] = []
245
+ entities: List[List[Dict]] = []
246
+
247
+ text_total_length = 0
248
+
249
+ po_start = 0
250
+
251
+ for i, p in enumerate(d.passages):
252
+
253
+ eo = p.offset - text_total_length
254
+
255
+ text_total_length += len(p.text) + 1
256
+
257
+ po_end = po_start + len(p.text)
258
+
259
+ dp = {
260
+ "text": p.text,
261
+ "type": p.infons.get("type"),
262
+ "offsets": [(po_start, po_end)],
263
+ "offset": p.offset, # original offset
264
+ }
265
+
266
+ po_start = po_end + 1
267
+
268
+ passages.append(dp)
269
+
270
+ pe = []
271
+
272
+ for a in p.annotations:
273
+
274
+ a_type = a.infons.get("type")
275
+
276
+ if (
277
+ self.config.schema == "bigbio_kb"
278
+ and a_type == "MeSH_Indexing_Chemical"
279
+ ):
280
+ continue
281
+
282
+ if (
283
+ a.text == None or a.text == ""
284
+ ) and self.config.schema == "bigbio_kb":
285
+ continue
286
+
287
+ offsets, text = get_texts_and_offsets_from_bioc_ann(a)
288
+
289
+ da = {
290
+ "type": a_type,
291
+ "offsets": [(start - eo, end - eo) for (start, end) in offsets],
292
+ "text": text,
293
+ "id": a.id,
294
+ "normalized": self._get_normalized(a),
295
+ }
296
+
297
+ pe.append(da)
298
+
299
+ entities.append(pe)
300
+
301
+ return passages, entities
302
+
303
+ def _get_normalized(self, a: bioc.BioCAnnotation) -> List[Dict]:
304
+ """
305
+ Get normalization DB and ID from annotation identifiers
306
+ """
307
+
308
+ identifiers = a.infons.get("identifier")
309
+
310
+ if identifiers is not None:
311
+
312
+ identifiers = re.split(r",|;", identifiers)
313
+
314
+ identifiers = [i for i in identifiers if i != "-"]
315
+
316
+ normalized = [i.split(":") for i in identifiers]
317
+
318
+ normalized = [
319
+ {"db_name": elems[0], "db_id": elems[1]} for elems in normalized
320
+ ]
321
+
322
+ else:
323
+
324
+ # No normalization
325
+ normalized = []
326
+
327
+ return normalized
328
+
329
+ def _get_textcls_example(self, d: bioc.BioCDocument) -> Dict:
330
+
331
+ example = {"document_id": d.id, "text": [], "labels": []}
332
+
333
+ for p in d.passages:
334
+
335
+ example["text"].append(p.text)
336
+
337
+ for a in p.annotations:
338
+
339
+ if a.infons.get("type") == "MeSH_Indexing_Chemical":
340
+
341
+ example["labels"].append(a.infons.get("identifier"))
342
+
343
+ example["text"] = " ".join(example["text"])
344
+
345
+ return example
346
+
347
+ def _generate_examples(
348
+ self,
349
+ filepath: str,
350
+ split: str,
351
+ ) -> Iterator[Tuple[int, Dict]]:
352
+ """Yields examples as (key, example) tuples."""
353
+
354
+ reader = biocxml.BioCXMLDocumentReader(str(filepath))
355
+
356
+ if self.config.schema == "source":
357
+
358
+ for uid, doc in enumerate(reader):
359
+
360
+ passages, passages_entities = self._get_passages_and_entities(doc)
361
+
362
+ for p, pe in zip(passages, passages_entities):
363
+
364
+ p.pop("offsets") # BioC has only start for passages offsets
365
+
366
+ p["document_id"] = doc.id
367
+ p["entities"] = pe # BioC has per passage entities
368
+
369
+ yield uid, {"passages": passages}
370
+
371
+ elif self.config.schema == "bigbio_kb":
372
+
373
+ uid = 0
374
+
375
+ for idx, doc in enumerate(reader):
376
+
377
+ passages, passages_entities = self._get_passages_and_entities(doc)
378
+
379
+ # global id
380
+ uid += 1
381
+
382
+ # unpack per-passage entities
383
+ entities = [e for pe in passages_entities for e in pe]
384
+
385
+ for p in passages:
386
+ p.pop("offset") # drop original offset
387
+ p["text"] = (p["text"],) # text in passage is Sequence
388
+ p["id"] = uid # override BioC default id
389
+ uid += 1
390
+
391
+ for e in entities:
392
+ e["id"] = uid # override BioC default id
393
+ uid += 1
394
+
395
+ yield idx, {
396
+ "id": uid,
397
+ "document_id": doc.id,
398
+ "passages": passages,
399
+ "entities": entities,
400
+ "events": [],
401
+ "coreferences": [],
402
+ "relations": [],
403
+ }
404
+
405
+ elif self.config.schema == "bigbio_text":
406
+
407
+ uid = 0
408
+
409
+ for idx, doc in enumerate(reader):
410
+
411
+ example = self._get_textcls_example(doc)
412
+ example["id"] = uid
413
+
414
+ # global id
415
+ uid += 1
416
+
417
+ yield idx, example