Spaces:
Sleeping
Sleeping
wzkariampuzha
commited on
Commit
•
aa32937
1
Parent(s):
f21365b
Upload classify_abs.py
Browse files- classify_abs.py +356 -0
classify_abs.py
ADDED
@@ -0,0 +1,356 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import requests
|
3 |
+
import xml.etree.ElementTree as ET
|
4 |
+
import pickle
|
5 |
+
import re
|
6 |
+
import os
|
7 |
+
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
|
8 |
+
import tensorflow as tf
|
9 |
+
from nltk.corpus import stopwords
|
10 |
+
from nltk.tokenize import word_tokenize
|
11 |
+
import spacy
|
12 |
+
import numpy as np
|
13 |
+
from tensorflow.keras.preprocessing.sequence import pad_sequences
|
14 |
+
STOPWORDS = set(stopwords.words('english'))
|
15 |
+
max_length = 300
|
16 |
+
trunc_type = 'post'
|
17 |
+
padding_type = 'post'
|
18 |
+
|
19 |
+
from typing import (
|
20 |
+
Dict,
|
21 |
+
List,
|
22 |
+
Tuple,
|
23 |
+
Set,
|
24 |
+
Optional,
|
25 |
+
Any,
|
26 |
+
Union,
|
27 |
+
)
|
28 |
+
|
29 |
+
# Standardize the abstract by replacing all named entities with their entity label.
|
30 |
+
# Eg. 3 patients reported at a clinic in England --> CARDINAL patients reported at a clinic in GPE
|
31 |
+
# expects the spaCy model en_core_web_lg as input
|
32 |
+
def standardizeAbstract(abstract:str, nlp:Any) -> str:
|
33 |
+
doc = nlp(abstract)
|
34 |
+
newAbstract = abstract
|
35 |
+
for e in reversed(doc.ents):
|
36 |
+
if e.label_ in {'PERCENT','CARDINAL','GPE','LOC','DATE','TIME','QUANTITY','ORDINAL'}:
|
37 |
+
start = e.start_char
|
38 |
+
end = start + len(e.text)
|
39 |
+
newAbstract = newAbstract[:start] + e.label_ + newAbstract[end:]
|
40 |
+
return newAbstract
|
41 |
+
|
42 |
+
# Same as above but replaces biomedical named entities from scispaCy models
|
43 |
+
# Expects as input en_ner_bc5cdr_md and en_ner_bionlp13cg_md
|
44 |
+
def standardizeSciTerms(abstract:str, nlpSci:Any, nlpSci2:Any) -> str:
|
45 |
+
doc = nlpSci(abstract)
|
46 |
+
newAbstract = abstract
|
47 |
+
for e in reversed(doc.ents):
|
48 |
+
start = e.start_char
|
49 |
+
end = start + len(e.text)
|
50 |
+
newAbstract = newAbstract[:start] + e.label_ + newAbstract[end:]
|
51 |
+
|
52 |
+
doc = nlpSci2(newAbstract)
|
53 |
+
for e in reversed(doc.ents):
|
54 |
+
start = e.start_char
|
55 |
+
end = start + len(e.text)
|
56 |
+
newAbstract = newAbstract[:start] + e.label_ + newAbstract[end:]
|
57 |
+
return newAbstract
|
58 |
+
|
59 |
+
# Prepare model
|
60 |
+
#nlp, nlpSci, nlpSci2, classify_model, classify_tokenizer= init_classify_model()
|
61 |
+
def init_classify_model(model:str='my_model_orphanet_final') -> Tuple[Any,Any,Any,Any,Any]:
|
62 |
+
#Load spaCy models
|
63 |
+
nlp = spacy.load('en_core_web_lg')
|
64 |
+
nlpSci = spacy.load("en_ner_bc5cdr_md")
|
65 |
+
nlpSci2 = spacy.load('en_ner_bionlp13cg_md')
|
66 |
+
|
67 |
+
# load the tokenizer
|
68 |
+
with open('tokenizer.pickle', 'rb') as handle:
|
69 |
+
classify_tokenizer = pickle.load(handle)
|
70 |
+
|
71 |
+
# load the model
|
72 |
+
classify_model = tf.keras.models.load_model(model)
|
73 |
+
|
74 |
+
return (nlp, nlpSci, nlpSci2, classify_model, classify_tokenizer)
|
75 |
+
|
76 |
+
#Gets abstract and title (concatenated) from EBI API
|
77 |
+
def PMID_getAb(PMID:Union[int,str]) -> str:
|
78 |
+
url = 'https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:'+str(PMID)+'&resulttype=core'
|
79 |
+
r = requests.get(url)
|
80 |
+
root = ET.fromstring(r.content)
|
81 |
+
titles = [title.text for title in root.iter('title')]
|
82 |
+
abstracts = [abstract.text for abstract in root.iter('abstractText')]
|
83 |
+
if len(abstracts) > 0 and len(abstracts[0])>5:
|
84 |
+
return titles[0]+' '+abstracts[0]
|
85 |
+
else:
|
86 |
+
return ''
|
87 |
+
|
88 |
+
def search_Pubmed_API(searchterm_list:Union[List[str],str], maxResults:int) -> Dict[str,str]: #returns a dictionary of {pmids:abstracts}
|
89 |
+
print('search_Pubmed_API is DEPRECATED. UTILIZE search_NCBI_API for NCBI ENTREZ API results. Utilize search_getAbs for most comprehensive results.')
|
90 |
+
return search_NCBI_API(searchterm_list, maxResults)
|
91 |
+
|
92 |
+
## DEPRECATED, use search_getAbs for more comprehensive results
|
93 |
+
def search_NCBI_API(searchterm_list:Union[List[str],str], maxResults:int) -> Dict[str,str]: #returns a dictionary of {pmids:abstracts}
|
94 |
+
print('search_NCBI_API is DEPRECATED. Utilize search_getAbs for most comprehensive results.')
|
95 |
+
pmid_to_abs = {}
|
96 |
+
i = 0
|
97 |
+
|
98 |
+
#type validation, allows string or list input
|
99 |
+
if type(searchterm_list)!=list:
|
100 |
+
if type(searchterm_list)==str:
|
101 |
+
searchterm_list = [searchterm_list]
|
102 |
+
else:
|
103 |
+
searchterm_list = list(searchterm_list)
|
104 |
+
|
105 |
+
#gathers pmids into a set first
|
106 |
+
for dz in searchterm_list:
|
107 |
+
# get results from searching for disease name through PubMed API
|
108 |
+
term = ''
|
109 |
+
dz_words = dz.split()
|
110 |
+
for word in dz_words:
|
111 |
+
term += word + '%20'
|
112 |
+
query = term[:-3]
|
113 |
+
url = 'https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term='+query
|
114 |
+
r = requests.get(url)
|
115 |
+
root = ET.fromstring(r.content)
|
116 |
+
|
117 |
+
# loop over resulting articles
|
118 |
+
for result in root.iter('IdList'):
|
119 |
+
pmids = [pmid.text for pmid in result.iter('Id')]
|
120 |
+
if i >= maxResults:
|
121 |
+
break
|
122 |
+
for pmid in pmids:
|
123 |
+
if pmid not in pmid_to_abs.keys():
|
124 |
+
abstract = PMID_getAb(pmid)
|
125 |
+
if len(abstract)>5:
|
126 |
+
pmid_to_abs[pmid]=abstract
|
127 |
+
i+=1
|
128 |
+
|
129 |
+
return pmid_to_abs
|
130 |
+
|
131 |
+
## DEPRECATED, use search_getAbs for more comprehensive results
|
132 |
+
# get results from searching for disease name through EBI API
|
133 |
+
def search_EBI_API(searchterm_list:Union[List[str],str], maxResults:int) -> Dict[str,str]: #returns a dictionary of {pmids:abstracts}
|
134 |
+
print('DEPRECATED. Utilize search_getAbs for most comprehensive results.')
|
135 |
+
pmids_abs = {}
|
136 |
+
i = 0
|
137 |
+
|
138 |
+
#type validation, allows string or list input
|
139 |
+
if type(searchterm_list)!=list:
|
140 |
+
if type(searchterm_list)==str:
|
141 |
+
searchterm_list = [searchterm_list]
|
142 |
+
else:
|
143 |
+
searchterm_list = list(searchterm_list)
|
144 |
+
|
145 |
+
#gathers pmids into a set first
|
146 |
+
for dz in searchterm_list:
|
147 |
+
if i >= maxResults:
|
148 |
+
break
|
149 |
+
term = ''
|
150 |
+
dz_words = dz.split()
|
151 |
+
for word in dz_words:
|
152 |
+
term += word + '%20'
|
153 |
+
query = term[:-3]
|
154 |
+
url = 'https://www.ebi.ac.uk/europepmc/webservices/rest/search?query='+query+'&resulttype=core'
|
155 |
+
r = requests.get(url)
|
156 |
+
root = ET.fromstring(r.content)
|
157 |
+
|
158 |
+
# loop over resulting articles
|
159 |
+
for result in root.iter('result'):
|
160 |
+
if i >= maxResults:
|
161 |
+
break
|
162 |
+
pmids = [pmid.text for pmid in result.iter('id')]
|
163 |
+
if len(pmids) > 0:
|
164 |
+
pmid = pmids[0]
|
165 |
+
if pmid[0].isdigit():
|
166 |
+
abstracts = [abstract.text for abstract in result.iter('abstractText')]
|
167 |
+
titles = [title.text for title in result.iter('title')]
|
168 |
+
if len(abstracts) > 0:# and len(abstracts[0])>5:
|
169 |
+
pmids_abs[pmid] = titles[0]+' '+abstracts[0]
|
170 |
+
i+=1
|
171 |
+
|
172 |
+
return pmids_abs
|
173 |
+
|
174 |
+
## This is the main, most comprehensive search_term function, it can take in a search term or a list of search terms and output a dictionary of {pmids:abstracts}
|
175 |
+
## Gets results from searching through both PubMed and EBI search term APIs, also makes use of the EBI API for PMIDs.
|
176 |
+
## EBI API and PubMed API give different results
|
177 |
+
# This makes n+2 API calls where n<=maxResults, which is slow
|
178 |
+
# There is a way to optimize by gathering abstracts from the EBI API when also getting pmids but did not pursue due to time constraints
|
179 |
+
# Filtering can be
|
180 |
+
# 'strict' - must have some exact match to at leastone of search terms/phrases in text)
|
181 |
+
# 'lenient' - part of the abstract must match at least one word in the search term phrases.
|
182 |
+
# 'none'
|
183 |
+
def search_getAbs(searchterm_list:Union[List[str],List[int],str], maxResults:int, filtering:str) -> Dict[str,str]:
|
184 |
+
#set of all pmids
|
185 |
+
pmids = set()
|
186 |
+
|
187 |
+
#dictionary {pmid:abstract}
|
188 |
+
pmid_abs = {}
|
189 |
+
|
190 |
+
#type validation, allows string or list input
|
191 |
+
if type(searchterm_list)!=list:
|
192 |
+
if type(searchterm_list)==str:
|
193 |
+
searchterm_list = [searchterm_list]
|
194 |
+
else:
|
195 |
+
searchterm_list = list(searchterm_list)
|
196 |
+
|
197 |
+
#gathers pmids into a set first
|
198 |
+
for dz in searchterm_list:
|
199 |
+
term = ''
|
200 |
+
dz_words = dz.split()
|
201 |
+
for word in dz_words:
|
202 |
+
term += word + '%20'
|
203 |
+
query = term[:-3]
|
204 |
+
|
205 |
+
## get pmid results from searching for disease name through PubMed API
|
206 |
+
url = 'https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term='+query
|
207 |
+
r = requests.get(url)
|
208 |
+
root = ET.fromstring(r.content)
|
209 |
+
|
210 |
+
# loop over resulting articles
|
211 |
+
for result in root.iter('IdList'):
|
212 |
+
if len(pmids) >= maxResults:
|
213 |
+
break
|
214 |
+
pmidlist = [pmid.text for pmid in result.iter('Id')]
|
215 |
+
pmids.update(pmidlist)
|
216 |
+
|
217 |
+
## get results from searching for disease name through EBI API
|
218 |
+
url = 'https://www.ebi.ac.uk/europepmc/webservices/rest/search?query='+query+'&resulttype=core'
|
219 |
+
r = requests.get(url)
|
220 |
+
root = ET.fromstring(r.content)
|
221 |
+
|
222 |
+
# loop over resulting articles
|
223 |
+
for result in root.iter('result'):
|
224 |
+
if len(pmids) >= maxResults:
|
225 |
+
break
|
226 |
+
pmidlist = [pmid.text for pmid in result.iter('id')]
|
227 |
+
#can also gather abstract and title here but for some reason did not work as intended the first time. Optimize in future versions to reduce latency.
|
228 |
+
if len(pmidlist) > 0:
|
229 |
+
pmid = pmidlist[0]
|
230 |
+
if pmid[0].isdigit():
|
231 |
+
pmids.add(pmid)
|
232 |
+
|
233 |
+
#Construct sets for filtering (right before adding abstract to pmid_abs
|
234 |
+
# The purpose of this is to do a second check of the abstracts, filters out any abstracts unrelated to the search terms
|
235 |
+
#if filtering is 'lenient' or default
|
236 |
+
if filtering !='none' or filtering !='strict':
|
237 |
+
filter_terms = set(searchterm_list).union(set(str(re.sub(',','',' '.join(searchterm_list))).split()).difference(STOPWORDS))
|
238 |
+
'''
|
239 |
+
# The above is equivalent to this but uses less memory and may be faster:
|
240 |
+
#create a single string of the terms within the searchterm_list
|
241 |
+
joined = ' '.join(searchterm_list)
|
242 |
+
#remove commas
|
243 |
+
comma_gone = re.sub(',','',joined)
|
244 |
+
#split the string into list of words and convert list into a Pythonic set
|
245 |
+
split = set(comma_gone.split())
|
246 |
+
#remove the STOPWORDS from the set of key words
|
247 |
+
key_words = split.difference(STOPWORDS)
|
248 |
+
#create a new set of the list members in searchterm_list
|
249 |
+
search_set = set(searchterm_list)
|
250 |
+
#join the two sets
|
251 |
+
terms = search_set.union(key_words)
|
252 |
+
#if any word(s) in the abstract intersect with any of these terms then the abstract is good to go.
|
253 |
+
'''
|
254 |
+
|
255 |
+
## get abstracts from EBI PMID API and output a dictionary
|
256 |
+
for pmid in pmids:
|
257 |
+
abstract = PMID_getAb(pmid)
|
258 |
+
if len(abstract)>5:
|
259 |
+
#do filtering here
|
260 |
+
if filtering == 'strict':
|
261 |
+
uncased_ab = abstract.lower()
|
262 |
+
for term in searchterm_list:
|
263 |
+
if term.lower() in uncased_ab:
|
264 |
+
pmid_abs[pmid] = abstract
|
265 |
+
break
|
266 |
+
elif filtering =='none':
|
267 |
+
pmid_abs[pmid] = abstract
|
268 |
+
|
269 |
+
#Default filtering is 'lenient'.
|
270 |
+
else:
|
271 |
+
#Else and if are separated for readability and to better understand logical flow.
|
272 |
+
if set(filter_terms).intersection(set(word_tokenize(abstract))):
|
273 |
+
pmid_abs[pmid] = abstract
|
274 |
+
|
275 |
+
|
276 |
+
print('Found',len(pmids),'PMIDs. Gathered',len(pmid_abs),'Relevant Abstracts.')
|
277 |
+
|
278 |
+
return pmid_abs
|
279 |
+
|
280 |
+
# Generate predictions for a PubMed Id
|
281 |
+
# nlp: en_core_web_lg
|
282 |
+
# nlpSci: en_ner_bc5cdr_md
|
283 |
+
# nlpSci2: en_ner_bionlp13cg_md
|
284 |
+
# Defaults to load my_model_orphanet_final, the most up-to-date version of the classification model,
|
285 |
+
# but can also be run on any other tf.keras model
|
286 |
+
#This was originally getPredictions
|
287 |
+
def getPMIDPredictions(pmid:Union[str,int], classify_model_vars:Tuple[Any,Any,Any,Any,Any]) -> Tuple[str,float,bool]:
|
288 |
+
nlp, nlpSci, nlpSci2, classify_model, classify_tokenizer = classify_model_vars
|
289 |
+
abstract = PMID_getAb(pmid)
|
290 |
+
|
291 |
+
if len(abstract)>5:
|
292 |
+
# remove stopwords
|
293 |
+
for word in STOPWORDS:
|
294 |
+
token = ' ' + word + ' '
|
295 |
+
abstract = abstract.replace(token, ' ')
|
296 |
+
abstract = abstract.replace(' ', ' ')
|
297 |
+
|
298 |
+
# preprocess abstract
|
299 |
+
abstract_standard = [standardizeAbstract(standardizeSciTerms(abstract, nlpSci, nlpSci2), nlp)]
|
300 |
+
sequence = classify_tokenizer.texts_to_sequences(abstract_standard)
|
301 |
+
padded = pad_sequences(sequence, maxlen=max_length, padding=padding_type, truncating=trunc_type)
|
302 |
+
|
303 |
+
y_pred1 = classify_model.predict(padded) # generate prediction
|
304 |
+
y_pred = np.argmax(y_pred1, axis=1) # get binary prediction
|
305 |
+
|
306 |
+
prob = y_pred1[0][1]
|
307 |
+
if y_pred == 1:
|
308 |
+
isEpi = True
|
309 |
+
else:
|
310 |
+
isEpi = False
|
311 |
+
|
312 |
+
return abstract, prob, isEpi
|
313 |
+
|
314 |
+
else:
|
315 |
+
return abstract, 0.0, False
|
316 |
+
|
317 |
+
|
318 |
+
def getTextPredictions(abstract:str, classify_model_vars:Tuple[Any,Any,Any,Any,Any]) -> Tuple[float,bool]:
|
319 |
+
|
320 |
+
nlp, nlpSci, nlpSci2, classify_model, classify_tokenizer = classify_model_vars
|
321 |
+
|
322 |
+
if len(abstract)>5:
|
323 |
+
# remove stopwords
|
324 |
+
for word in STOPWORDS:
|
325 |
+
token = ' ' + word + ' '
|
326 |
+
abstract = abstract.replace(token, ' ')
|
327 |
+
abstract = abstract.replace(' ', ' ')
|
328 |
+
|
329 |
+
# preprocess abstract
|
330 |
+
abstract_standard = [standardizeAbstract(standardizeSciTerms(abstract, nlpSci, nlpSci2), nlp)]
|
331 |
+
sequence = classify_tokenizer.texts_to_sequences(abstract_standard)
|
332 |
+
padded = pad_sequences(sequence, maxlen=max_length, padding=padding_type, truncating=trunc_type)
|
333 |
+
|
334 |
+
y_pred1 = classify_model.predict(padded) # generate prediction
|
335 |
+
y_pred = np.argmax(y_pred1, axis=1) # get binary prediction
|
336 |
+
|
337 |
+
prob = y_pred1[0][1]
|
338 |
+
if y_pred == 1:
|
339 |
+
isEpi = True
|
340 |
+
else:
|
341 |
+
isEpi = False
|
342 |
+
|
343 |
+
return prob, isEpi
|
344 |
+
|
345 |
+
else:
|
346 |
+
return 0.0, False
|
347 |
+
|
348 |
+
if __name__ == '__main__':
|
349 |
+
print('Loading 5 NLP models...')
|
350 |
+
classify_model_vars= init_classify_model()
|
351 |
+
print('All models loaded.')
|
352 |
+
pmid = input('\nEnter PubMed PMID (or DONE): ')
|
353 |
+
while pmid != 'DONE':
|
354 |
+
abstract, prob, isEpi = getPredictions(pmid, classify_model_vars)
|
355 |
+
print(abstract, prob, isEpi)
|
356 |
+
pmid = input('\nEnter PubMed PMID (or DONE): ')
|