Description
Extract potential improvements or harmful effects of Drug-Drug interactions (DDIs) when two or more drugs are taken at the same time or at a certain interval.
Predicted Entities
DDI-advise, DDI-effect, DDI-false, DDI-int, DDI-mechanism
Live Demo Open in Colab Copy S3 URI
How to use
...
documenter = DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
sentencer = SentenceDetector()\
.setInputCols(["document"])\
.setOutputCol("sentences")
tokenizer = sparknlp.annotators.Tokenizer()\
.setInputCols(["sentences"])\
.setOutputCol("tokens")
pos_tagger = PerceptronModel()\
.pretrained("pos_clinical", "en", "clinical/models") \
.setInputCols(["sentences", "tokens"])\
.setOutputCol("pos_tags")
words_embedder = WordEmbeddingsModel() \
.pretrained("embeddings_clinical", "en", "clinical/models") \
.setInputCols(["sentences", "tokens"]) \
.setOutputCol("embeddings")
ner_tagger = MedicalNerModel.pretrained("ner_posology", "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")
# Set a filter on pairs of named entities which will be treated as relation candidates
re_ner_chunk_filter = RENerChunksFilter() \
.setInputCols(["ner_chunks", "dependencies"])\
.setMaxSyntacticDistance(10)\
.setOutputCol("re_ner_chunks")
#.setRelationPairs(['SYMPTOM-EXTERNAL_BODY_PART_OR_REGION'])
# The dataset this model is trained to is sentence-wise.
# This model can also be trained on document-level relations - in which case, while predicting, use "document" instead of "sentence" as input.
re_model = RelationExtractionDLModel()\
.pretrained('redl_drug_drug_interaction_biobert', 'en', "clinical/models") \
.setPredictionThreshold(0.5)\
.setInputCols(["re_ner_chunks", "sentences"]) \
.setOutputCol("relations")
pipeline = Pipeline(stages=[documenter, sentencer, tokenizer, pos_tagger, words_embedder, ner_tagger, ner_converter, dependency_parser, re_ner_chunk_filter, re_model])
text="""When carbamazepine is withdrawn from the combination therapy, aripiprazole dose should then be reduced.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"""
data = spark.createDataFrame([[text]]).toDF("text")
p_model = pipeline.fit(data)
result = p_model.transform(data)
...
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 pos_tagger = PerceptronModel()
.pretrained("pos_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens"))
.setOutputCol("pos_tags")
val words_embedder = WordEmbeddingsModel()
.pretrained("embeddings_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens"))
.setOutputCol("embeddings")
val ner_tagger = MedicalNerModel.pretrained("ner_posology", "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")
// Set a filter on pairs of named entities which will be treated as relation candidates
val re_ner_chunk_filter = RENerChunksFilter()
.setInputCols(Array("ner_chunks", "dependencies"))
.setMaxSyntacticDistance(10)
.setOutputCol("re_ner_chunks")
// .setRelationPairs(Array('SYMPTOM-EXTERNAL_BODY_PART_OR_REGION'))
// The dataset this model is trained to is sentence-wise.
// This model can also be trained on document-level relations - in which case, while predicting, use "document" instead of "sentence" as input.
val re_model = RelationExtractionDLModel()
.pretrained("redl_drug_drug_interaction_biobert", "en", "clinical/models")
.setPredictionThreshold(0.5)
.setInputCols(Array("re_ner_chunks", "sentences"))
.setOutputCol("relations")
val pipeline = new Pipeline().setStages(Array(documenter, sentencer, tokenizer, pos_tagger, words_embedder, ner_tagger, ner_converter, dependency_parser, re_ner_chunk_filter, 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)
import nlu
nlu.load("en.relation.drug_drug_interaction").predict("""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""")
Results
+---------+-------+-------------+-----------+-------------+-------+-------------+-----------+------------+----------+
| relation|entity1|entity1_begin|entity1_end| chunk1|entity2|entity2_begin|entity2_end| chunk2|confidence|
+---------+-------+-------------+-----------+-------------+-------+-------------+-----------+------------+----------+
|DDI-false| DRUG| 5| 17|carbamazepine| DRUG| 62| 73|aripiprazole|0.91685396|
+---------+-------+-------------+-----------+-------------+-------+-------------+-----------+------------+----------+
Model Information
| Model Name: | redl_drug_drug_interaction_biobert |
| Compatibility: | Healthcare NLP 3.0.3+ |
| License: | Licensed |
| Edition: | Official |
| Language: | en |
| Case sensitive: | true |
Data Source
Trained on DDI Extraction corpus.
Benchmarking
Relation Recall Precision F1 Support
DDI-advise 0.758 0.874 0.812 211
DDI-effect 0.759 0.754 0.756 348
DDI-false 0.977 0.957 0.967 4097
DDI-int 0.175 0.458 0.253 63
DDI-mechanism 0.783 0.853 0.816 281
Avg. 0.690 0.779 0.721 -