Datasets:

Languages:
English
License:
gabrielaltay commited on
Commit
024bcc2
1 Parent(s): f500cf2

upload hubscripts/psytar_hub.py to hub from bigbio repo

Browse files
Files changed (1) hide show
  1. psytar.py +506 -0
psytar.py ADDED
@@ -0,0 +1,506 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ """
17
+ The "Psychiatric Treatment Adverse Reactions" (PsyTAR) dataset contains 891 drugs
18
+ reviews posted by patients on "askapatient.com", about the effectiveness and adverse
19
+ drug events associated with Zoloft, Lexapro, Cymbalta, and Effexor XR.
20
+
21
+ For each drug review, patient demographics, duration of treatment, and satisfaction
22
+ with the drugs were reported.
23
+
24
+ This dataset can be used for:
25
+
26
+ 1. (multi-label) sentence classification, across 5 labels:
27
+ Adverse Drug Reaction (ADR)
28
+ Withdrawal Symptoms (WDs)
29
+ Sign/Symptoms/Illness (SSIs)
30
+ Drug Indications (DIs)
31
+ Drug Effectiveness (EF)
32
+ Drug Infectiveness (INF)
33
+ and Others (not applicable)
34
+
35
+ 2. Recognition of 5 different types of entity:
36
+ ADRs (4813 mentions)
37
+ WDs (590 mentions)
38
+ SSIs (1219 mentions)
39
+ DIs (792 mentions)
40
+
41
+ In the source schema, systematic annotation with UMLS and SNOMED-CT concepts are provided.
42
+ """
43
+
44
+ import re
45
+ from dataclasses import dataclass
46
+ from pathlib import Path
47
+ from typing import Dict, List, Tuple
48
+
49
+ import datasets
50
+ import pandas as pd
51
+
52
+ from .bigbiohub import kb_features
53
+ from .bigbiohub import BigBioConfig
54
+ from .bigbiohub import Tasks
55
+
56
+ _LANGUAGES = ['English']
57
+ _PUBMED = False
58
+ _LOCAL = True
59
+ _CITATION = """\
60
+ @article{Zolnoori2019,
61
+ author = {Maryam Zolnoori and
62
+ Kin Wah Fung and
63
+ Timothy B. Patrick and
64
+ Paul Fontelo and
65
+ Hadi Kharrazi and
66
+ Anthony Faiola and
67
+ Yi Shuan Shirley Wu and
68
+ Christina E. Eldredge and
69
+ Jake Luo and
70
+ Mike Conway and
71
+ Jiaxi Zhu and
72
+ Soo Kyung Park and
73
+ Kelly Xu and
74
+ Hamideh Moayyed and
75
+ Somaieh Goudarzvand},
76
+ title = {A systematic approach for developing a corpus of patient \
77
+ reported adverse drug events: A case study for {SSRI} and {SNRI} medications},
78
+ journal = {Journal of Biomedical Informatics},
79
+ volume = {90},
80
+ year = {2019},
81
+ url = {https://doi.org/10.1016/j.jbi.2018.12.005},
82
+ doi = {10.1016/j.jbi.2018.12.005},
83
+ }
84
+ """
85
+
86
+ _DATASETNAME = "psytar"
87
+ _DISPLAYNAME = "PsyTAR"
88
+
89
+ _DESCRIPTION = """\
90
+ The "Psychiatric Treatment Adverse Reactions" (PsyTAR) dataset contains 891 drugs
91
+ reviews posted by patients on "askapatient.com", about the effectiveness and adverse
92
+ drug events associated with Zoloft, Lexapro, Cymbalta, and Effexor XR.
93
+
94
+ This dataset can be used for (multi-label) sentence classification of Adverse Drug
95
+ Reaction (ADR), Withdrawal Symptoms (WDs), Sign/Symptoms/Illness (SSIs), Drug
96
+ Indications (DIs), Drug Effectiveness (EF), Drug Infectiveness (INF) and Others, as well
97
+ as for recognition of 5 different types of named entity (in the categories ADRs, WDs,
98
+ SSIs and DIs)
99
+ """
100
+
101
+ _HOMEPAGE = "https://www.askapatient.com/research/pharmacovigilance/corpus-ades-psychiatric-medications.asp"
102
+
103
+ _LICENSE = 'Creative Commons Attribution 4.0 International'
104
+
105
+ _SUPPORTED_TASKS = [Tasks.NAMED_ENTITY_RECOGNITION, Tasks.TEXT_CLASSIFICATION]
106
+
107
+ _SOURCE_VERSION = "1.0.0"
108
+ _BIGBIO_VERSION = "1.0.0"
109
+
110
+
111
+ @dataclass
112
+ class PsyTARBigBioConfig(BigBioConfig):
113
+ schema: str = "source"
114
+ name: str = "psytar_source"
115
+ version: datasets.Version = _SOURCE_VERSION
116
+ description: str = "PsyTAR source schema"
117
+ subset_id: str = "psytar"
118
+
119
+
120
+ class PsyTARDataset(datasets.GeneratorBasedBuilder):
121
+ """The PsyTAR dataset contains patient's reviews on the effectiveness and adverse
122
+ drug events associated with Zoloft, Lexapro, Cymbalta, and Effexor XR."""
123
+
124
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
125
+ BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
126
+
127
+ BUILDER_CONFIGS = [
128
+ PsyTARBigBioConfig(
129
+ name="psytar_source",
130
+ version=SOURCE_VERSION,
131
+ description="PsyTAR source schema",
132
+ schema="source",
133
+ subset_id="psytar",
134
+ ),
135
+ PsyTARBigBioConfig(
136
+ name="psytar_bigbio_kb",
137
+ version=BIGBIO_VERSION,
138
+ description="PsyTAR BigBio KB schema",
139
+ schema="bigbio_kb",
140
+ subset_id="psytar",
141
+ ),
142
+ PsyTARBigBioConfig(
143
+ name="psytar_bigbio_text",
144
+ version=BIGBIO_VERSION,
145
+ description="PsyTAR BigBio text classification schema",
146
+ schema="bigbio_text",
147
+ subset_id="psytar",
148
+ ),
149
+ ]
150
+
151
+ BUILDER_CONFIG_CLASS = PsyTARBigBioConfig
152
+
153
+ DEFAULT_CONFIG_NAME = "psytar_source"
154
+
155
+ def _info(self) -> datasets.DatasetInfo:
156
+
157
+ if self.config.schema == "source":
158
+ features = datasets.Features(
159
+ {
160
+ "id": datasets.Value("string"),
161
+ "doc_id": datasets.Value("string"),
162
+ "disorder": datasets.Value("string"),
163
+ "side_effect": datasets.Value("string"),
164
+ "comment": datasets.Value("string"),
165
+ "gender": datasets.Value("string"),
166
+ "age": datasets.Value("int32"),
167
+ "dosage_duration": datasets.Value("string"),
168
+ "date": datasets.Value("string"),
169
+ "category": datasets.Value("string"),
170
+ "sentences": [
171
+ {
172
+ "text": datasets.Value("string"),
173
+ "label": datasets.Sequence([datasets.Value("string")]),
174
+ "findings": datasets.Value("string"),
175
+ "others": datasets.Value("string"),
176
+ "rating": datasets.Value("string"),
177
+ "category": datasets.Value("string"),
178
+ "entities": [
179
+ {
180
+ "text": datasets.Value("string"),
181
+ "type": datasets.Value("string"),
182
+ "mild": datasets.Value("string"),
183
+ "moderate": datasets.Value("string"),
184
+ "severe": datasets.Value("string"),
185
+ "persistent": datasets.Value("string"),
186
+ "non_persistent": datasets.Value("string"),
187
+ "body_site": datasets.Value("string"),
188
+ "rating": datasets.Value("string"),
189
+ "drug": datasets.Value("string"),
190
+ "class": datasets.Value("string"),
191
+ "entity_type": datasets.Value("string"),
192
+ "UMLS": datasets.Sequence(
193
+ [datasets.Value("string")]
194
+ ),
195
+ "SNOMED": datasets.Sequence(
196
+ [datasets.Value("string")]
197
+ ),
198
+ }
199
+ ],
200
+ }
201
+ ],
202
+ }
203
+ )
204
+ elif self.config.schema == "bigbio_kb":
205
+ features = kb_features
206
+ elif self.config.schema == "bigbio_text":
207
+ features = text_features
208
+
209
+ return datasets.DatasetInfo(
210
+ description=_DESCRIPTION,
211
+ features=features,
212
+ homepage=_HOMEPAGE,
213
+ license=str(_LICENSE),
214
+ citation=_CITATION,
215
+ )
216
+
217
+ def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]:
218
+ """Returns SplitGenerators."""
219
+ if self.config.data_dir is None:
220
+ raise ValueError(
221
+ "This is a local dataset. Please pass the data_dir kwarg to load_dataset."
222
+ )
223
+ else:
224
+ data_dir = self.config.data_dir
225
+
226
+ return [
227
+ datasets.SplitGenerator(
228
+ name=datasets.Split.TRAIN,
229
+ gen_kwargs={
230
+ "filepath": Path(data_dir),
231
+ },
232
+ ),
233
+ ]
234
+
235
+ def _extract_labels(self, row):
236
+ label = [
237
+ "ADR" * row.ADR,
238
+ "WD" * row.WD,
239
+ "EF" * row.EF,
240
+ "INF" * row.INF,
241
+ "SSI" * row.SSI,
242
+ "DI" * row.DI,
243
+ "Others" * row.others,
244
+ ]
245
+ label = [_l for _l in label if _l != ""]
246
+ return label
247
+
248
+ def _columns_to_list(self, row, sheet="ADR"):
249
+ annotations = []
250
+ for i in range(30 if sheet == "ADR" else 10):
251
+ annotations.append(row[f"{sheet}{i + 1}"])
252
+ annotations = [a for a in annotations if not pd.isna(a)]
253
+ return annotations
254
+
255
+ def _columns_to_bigbio_kb(self, row, sheet="ADR"):
256
+ annotations = []
257
+ for i in range(30 if sheet == "ADR" else 10):
258
+ annotation = row[f"{sheet}{i + 1}"]
259
+ if not pd.isna(annotation):
260
+ start_index = row.sentences.lower().find(annotation.lower())
261
+ if start_index != -1:
262
+ end_index = start_index + len(annotation)
263
+ entity = {
264
+ "id": f"T{i+1}",
265
+ "offsets": [[start_index, end_index]],
266
+ "text": [annotation],
267
+ "type": sheet,
268
+ }
269
+
270
+ annotations.append(entity)
271
+ return annotations
272
+
273
+ def _standards_columns_to_list(self, row, standard="UMLS"):
274
+ standards = {"UMLS": ["UMLS1", "UMLS2"], "SNOMED": ["SNOMED-CT", "SNOMED-CT.1"]}
275
+ _out_list = []
276
+ for s in standards[standard]:
277
+ _out_list.append(row[s])
278
+ _out_list = [a for a in _out_list if not pd.isna(a)]
279
+ return _out_list
280
+
281
+ def _read_sentence_xlsx(self, filepath: Path) -> pd.DataFrame:
282
+ sentence_df = pd.read_excel(
283
+ filepath,
284
+ sheet_name="Sentence_Labeling",
285
+ dtype={"drug_id": str, "sentences": str},
286
+ )
287
+
288
+ sentence_df = sentence_df.dropna(subset=["sentences"])
289
+ sentence_df = sentence_df.loc[
290
+ sentence_df.sentences.apply(lambda x: len(x.strip())) > 0
291
+ ]
292
+ sentence_df = sentence_df.fillna(0)
293
+
294
+ sentence_df[["ADR", "WD", "EF", "INF", "SSI", "DI"]] = (
295
+ sentence_df[["ADR", "WD", "EF", "INF", "SSI", "DI"]]
296
+ .replace(re.compile("[!* ]+"), 1)
297
+ .astype(int)
298
+ )
299
+
300
+ sentence_df["sentence_index"] = sentence_df["sentence_index"].astype("int32")
301
+ sentence_df["drug_id"] = sentence_df["drug_id"].astype("str")
302
+
303
+ return sentence_df
304
+
305
+ def _read_samples_xlsx(self, filepath: Path) -> pd.DataFrame:
306
+ samples_df = pd.read_excel(
307
+ filepath, sheet_name="Sample", dtype={"drug_id": str}
308
+ )
309
+ samples_df["age"] = samples_df["age"].fillna(0).astype(int)
310
+ samples_df["drug_id"] = samples_df["drug_id"].astype("str")
311
+
312
+ return samples_df
313
+
314
+ def _read_identified_xlsx_to_bigbio_kb(self, filepath: Path) -> Dict:
315
+ sheet_names = ["ADR", "WD", "SSI", "DI"]
316
+ identified_entities = {}
317
+
318
+ for sheet in sheet_names:
319
+ identified_entities[sheet] = pd.read_excel(
320
+ filepath, sheet_name=sheet + "_Identified"
321
+ )
322
+ identified_entities[sheet]["bigbio_kb"] = identified_entities[sheet].apply(
323
+ lambda x: self._columns_to_bigbio_kb(x, sheet), axis=1
324
+ )
325
+
326
+ return identified_entities
327
+
328
+ TYPE_TO_COLNAME = {"ADR": "ADRs", "DI": "DIs", "SSI": "SSI", "WD": "WDs"}
329
+
330
+ def _identified_mapped_xlsx_to_df(self, filepath: Path) -> pd.DataFrame:
331
+ sheet_names_mapped = [
332
+ ["ADR_Mapped", "ADR"],
333
+ ["WD-Mapped ", "WD"],
334
+ ["SSI_Mapped", "SSI"],
335
+ ["DI_Mapped", "DI"],
336
+ ]
337
+
338
+ _mappings = []
339
+
340
+ # Read the specific XLSX sheet with _Mapped annotations
341
+ for sheet, sheet_short in sheet_names_mapped:
342
+ _df_mapping = pd.read_excel(filepath, sheet_name=sheet)
343
+
344
+ # Correcting column names
345
+ if sheet_short in ["WD"]:
346
+ _df_mapping = _df_mapping.rename(
347
+ columns={"sentence_id": "sentence_index"}
348
+ )
349
+
350
+ # Changing column names to allow concatenation
351
+ _df_mapping = _df_mapping.rename(
352
+ columns={self.TYPE_TO_COLNAME[sheet_short]: "entity"}
353
+ )
354
+
355
+ # Putting UMLS and SNOMED annotations in a single column
356
+ _df_mapping["UMLS"] = _df_mapping.apply(
357
+ lambda x: self._standards_columns_to_list(x), axis=1
358
+ )
359
+ _df_mapping["SNOMED"] = _df_mapping.apply(
360
+ lambda x: self._standards_columns_to_list(x, standard="SNOMED"), axis=1
361
+ )
362
+
363
+ _mappings.append(_df_mapping)
364
+
365
+ df_mappings = pd.concat(_mappings).fillna(0)
366
+ df_mappings["sentence_index"] = df_mappings["sentence_index"].astype("int32")
367
+ df_mappings["drug_id"] = df_mappings["drug_id"].astype("str")
368
+
369
+ return df_mappings
370
+
371
+ def _convert_xlsx_to_source(self, filepath: Path) -> Dict:
372
+ # Read XLSX files
373
+ df_sentences = self._read_sentence_xlsx(filepath)
374
+ df_sentences["label"] = df_sentences.apply(
375
+ lambda x: self._extract_labels(x), axis=1
376
+ )
377
+ df_mappings = self._identified_mapped_xlsx_to_df(filepath)
378
+ df_samples = self._read_samples_xlsx(filepath)
379
+
380
+ # Configure indices
381
+ df_samples = df_samples.set_index("drug_id").sort_index()
382
+ df_sentences = df_sentences.set_index(
383
+ ["drug_id", "sentence_index"]
384
+ ).sort_index()
385
+ df_mappings = df_mappings.set_index(["drug_id", "sentence_index"]).sort_index()
386
+
387
+ # Iterate over samples
388
+ for sample_row_id, sample in df_samples.iterrows():
389
+ sentences = []
390
+ try:
391
+ df_sentence_selection = df_sentences.loc[sample_row_id]
392
+
393
+ # Iterate over sentences
394
+ for sentence_row_id, sentence in df_sentence_selection.iterrows():
395
+ entities = []
396
+ try:
397
+ df_mapped_selection = df_mappings.loc[
398
+ sample_row_id, sentence_row_id
399
+ ]
400
+
401
+ # Iterate over entities per sentence
402
+ for mapped_row_id, row in df_mapped_selection.iterrows():
403
+ entities.append(
404
+ {
405
+ "text": row["entity"],
406
+ "UMLS": row.UMLS,
407
+ "SNOMED": row.SNOMED,
408
+ "entity_type": row.entity_type,
409
+ "type": row.type,
410
+ "class": row["class"],
411
+ "drug": row.drug,
412
+ "rating": row.rating,
413
+ "body_site": row["body-site"],
414
+ "non_persistent": row["not-persistent"],
415
+ "persistent": row["persistent"],
416
+ "severe": row.severe,
417
+ "moderate": row.moderate,
418
+ "mild": row.mild,
419
+ }
420
+ )
421
+ except KeyError:
422
+ pass
423
+
424
+ sentences.append(
425
+ {
426
+ "text": sentence.sentences,
427
+ "entities": entities,
428
+ "label": sentence.label,
429
+ "findings": sentence.Findings,
430
+ "others": sentence.others,
431
+ "rating": sentence.rating,
432
+ "category": sentence.category,
433
+ }
434
+ )
435
+ except KeyError:
436
+ pass
437
+
438
+ example = {
439
+ "id": sample_row_id,
440
+ "doc_id": sample_row_id,
441
+ "disorder": sample.disorder,
442
+ "side_effect": sample["side-effect"],
443
+ "comment": sample.comment,
444
+ "gender": sample.gender,
445
+ "age": sample.age,
446
+ "dosage_duration": sample.dosage_duration,
447
+ "date": str(sample.date),
448
+ "category": sample.category,
449
+ "sentences": sentences,
450
+ }
451
+ yield example
452
+
453
+ def _convert_xlsx_to_bigbio_kb(self, filepath: Path) -> Dict:
454
+ bigbio_kb = self._read_identified_xlsx_to_bigbio_kb(filepath)
455
+
456
+ i_doc = 0
457
+ for _, df in bigbio_kb.items():
458
+ for _, row in df.iterrows():
459
+ text = row.sentences
460
+ entities = row["bigbio_kb"]
461
+ doc_id = f"{row['drug_id']}_{row['sentence_index']}_{i_doc}"
462
+
463
+ if len(entities) != 0:
464
+ example = parsing.brat_parse_to_bigbio_kb(
465
+ {
466
+ "document_id": doc_id,
467
+ "text": text,
468
+ "text_bound_annotations": entities,
469
+ "normalizations": [],
470
+ "events": [],
471
+ "relations": [],
472
+ "equivalences": [],
473
+ "attributes": [],
474
+ },
475
+ )
476
+ example["id"] = i_doc
477
+ i_doc += 1
478
+ yield example
479
+
480
+ def _convert_xlsx_to_bigbio_text(self, filepath: Path) -> Dict:
481
+ df = self._read_sentence_xlsx(filepath)
482
+ df["label"] = df.apply(lambda x: self._extract_labels(x), axis=1)
483
+
484
+ for idx, row in df.iterrows():
485
+ example = {
486
+ "id": idx,
487
+ "document_id": f"{row['drug_id']}_{row['sentence_index']}",
488
+ "text": row["label"],
489
+ "labels": row["category"],
490
+ }
491
+ yield example
492
+
493
+ def _generate_examples(self, filepath) -> Tuple[int, Dict]:
494
+ """Yields examples as (key, example) tuples."""
495
+
496
+ if self.config.schema == "source":
497
+ examples = self._convert_xlsx_to_source(filepath)
498
+
499
+ elif self.config.schema == "bigbio_kb":
500
+ examples = self._convert_xlsx_to_bigbio_kb(filepath)
501
+
502
+ elif self.config.schema == "bigbio_text":
503
+ examples = self._convert_xlsx_to_bigbio_text(filepath)
504
+
505
+ for idx, example in enumerate(examples):
506
+ yield idx, example