Description
Relation Extraction model based on syntactic features using deep learning. This model can be used to identify drug-drug interactions relationships among drug entities.
Predicted Entities
DDI-advise, DDI-effect, DDI-mechanism, DDI-int, DDI-false
How to use
In the table below, re_drug_drug_interaction_clinical RE model, its labels, optimal NER model, and meaningful relation pairs are illustrated.
| RE MODEL | RE MODEL LABES | NER MODEL | RE PAIRS |
|---|---|---|---|
| re_drug_drug_interaction_clinical | DDI-advise, DDI-effect, DDI-mechanism, DDI-int, DDI-false |
ner_posology | [“drug-drug”] |
documenter = DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
sentencer = SentenceDetector()\
.setInputCols(["document"])\
.setOutputCol("sentences")
tokenizer = Tokenizer()\
.setInputCols(["sentences"])\
.setOutputCol("tokens")
words_embedder = WordEmbeddingsModel.pretrained("embeddings_clinical", "en", "clinical/models")\
.setInputCols(["sentences", "tokens"])\
.setOutputCol("embeddings")
pos_tagger = PerceptronModel()\
.pretrained("pos_clinical", "en", "clinical/models") \
.setInputCols(["sentences", "tokens"])\
.setOutputCol("pos_tags")
ner_tagger = MedicalNerModel.pretrained("ner_clinical", "en", "clinical/models")\
.setInputCols("sentences", "tokens", "embeddings")\
.setOutputCol("ner_tags")
ner_converter = NerConverter()\
.setInputCols(["sentences", "tokens", "ner_tags"])\
.setOutputCol("ner_chunks")
dependency_parser = DependencyParserModel()\
.pretrained("dependency_conllu", "en")\
.setInputCols(["sentences", "pos_tags", "tokens"])\
.setOutputCol("dependencies")
ddi_re_model = RelationExtractionModel.pretrained("re_drug_drug_interaction_clinical","en","clinical/models")\
.setInputCols("word_embeddings","chunk","pos","dependency")\
.setOutputCol("category")
nlp_pipeline = Pipeline(stages=[documenter, sentencer, tokenizer, words_embedder, pos_tagger, ner_tagger, ner_converter, dependency_parser, ddi_re_model])
light_pipeline = LightPipeline(nlp_pipeline.fit(spark.createDataFrame([[""]]).toDF("text")))
annotations = light_pipeline.fullAnnotate("""When carbamazepine is withdrawn from the combination therapy, aripiprazole dose should then be reduced. If additional adrenergic drugs are to be administered by any route, they should be used with caution because the pharmacologically predictable sympathetic effects of Metformin may be potentiated""")
val documenter = new DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val sentencer = new SentenceDetector()
.setInputCols("document")
.setOutputCol("sentences")
val tokenizer = new Tokenizer()
.setInputCols("sentences")
.setOutputCol("tokens")
val words_embedder = WordEmbeddingsModel().pretrained("embeddings_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens"))
.setOutputCol("embeddings")
val pos_tagger = PerceptronModel().pretrained("pos_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens"))
.setOutputCol("pos_tags")
val ner_tagger = MedicalNerModel.pretrained("ner_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens", "embeddings"))
.setOutputCol("ner_tags")
val ner_converter = new NerConverter()
.setInputCols(Array("sentences", "tokens", "ner_tags"))
.setOutputCol("ner_chunks")
val dependency_parser = DependencyParserModel().pretrained("dependency_conllu", "en")
.setInputCols(Array("sentences", "pos_tags", "tokens"))
.setOutputCol("dependencies")
val ddi_re_model = RelationExtractionModel.pretrained("re_drug_drug_interaction_clinical","en","clinical/models")
.setInputCols("word_embeddings","chunk","pos","dependency")
.setOutputCol("category")
val pipeline = new Pipeline().setStages(Array(documenter, sentencer, tokenizer, words_embedder, pos_tagger, ner_tagger, ner_converter, dependency_parser, ddi_re_model))
val data = Seq("""When carbamazepine is withdrawn from the combination therapy, aripiprazole dose should then be reduced. If additional adrenergic drugs are to be administered by any route, they should be used with caution because the pharmacologically predictable sympathetic effects of Metformin may be potentiated""").toDS().toDF("text")
val result = pipeline.fit(data).transform(data)
Results
|relation | entity1 | entity1_begin | entity1_end | chunk1 | entity2 |entity2_begin | entity2_end | chunk2 |
|DDI-advise | DRUG | 5 | 17 | carbamazepine | DRUG | 62 73 | aripiprazole |
Model Information
| Name: | re_drug_drug_interaction_clinical | |
| Type: | RelationExtractionModel | |
| Compatibility: | Spark NLP 2.5.5+ | |
| License: | Licensed | |
| Edition: | Official | |
| Input labels: | [word_embeddings, chunk, pos, dependency] | |
| Output labels: | [category] | |
| Language: | en | |
| Case sensitive: | False | |
| Dependencies: | embeddings_clinical |
Data Source
Trained on data gathered and manually annotated by John Snow Labs.
Benchmarking
+-------------+------+------+------+
| relation|recall| prec | f1 |
+-------------+------+------+------+
| DDI-effect| 0.76| 0.38 | 0.51 |
| DDI-false| 0.72| 0.97 | 0.83 |
| DDI-advise| 0.74| 0.39 | 0.51 |
+-------------+------+------+------+