Harnessing the Power of NLP and Knowledge Graphs for Opioid Research

Gursev Pirge, PhD
Senior Data Scientist - JohnSnowLabs

1. Harnessing the Power of NLP and Knowledge
Graphs for Opioid Research
GursevPirge, PhD
Data Scientist
John Snow Labs

2. HealthcareNLP
Spark NLP for Healthcare provides
-accurate,
-scalable,
-private,
-tuneable,
-modular
software library that helps healthcare
& pharma organizations build
longitudinal patient records and
knowledge graphs on real-world EHR
data.

3. Data Origination and Exchange

5. Spark NLP for Healthcare
Named Entity
Recognition
ICD10 Resolver
Snomed
Resolver
UMLS Resolver
Risk Adj. Module RxNorm Resolver
Assertion Status
Detection
Sentence Splitter
Tokenizer
Bert Embeddings
Relationship
Extraction
sBert
Embeddings

6. Spark NLP for Healthcare
CT ABDOMEN:
There is no evidence
for a retroperitoneal
hematoma. Within the
superior pole of the
left kidney, there is
a3.9 cm cystic
lesion. A 3.3 cm
cystic lesion is also
seen within the
inferior pole of the
left kidney. No
calcifications are
noted.

7. Medical Question Answering
7

10. Opioids
-A class of drugs that include both natural and synthetic substances derived from or related to
the opium poppy plant.
-Act on opioid receptors in the brain and nervous system to provide pain relief, sedation, and
euphoria.
-Natural opiates (morphine, codeine), semi-synthetic opioids (hydrocodone, oxycodone, heroin),
and synthetic opioids (fentanyl, methadone, tramadol).
-Have high potential for misuse, addiction, and overdose, contributing to the ongoing opioid
epidemic.
10

11. Opioids
11
https://nida.nih.gov/research-
topics/trends-statistics/overdose-
death-rates

12. Opioids - NER

13. Opioids - NER
13

14. Opioids – Assertion Status Detection
14

15. Opioids – Assertion Status Detection
15

16. Opioids – RE
16

17. 17

18. 18

20. 20

21. 21

22. 22

23. 23

24. Neo4J – Demo App
-Collect data from PubMed - a free database including primarily the MEDLINE database of
references and abstracts on life sciences and biomedical topics.
-Use pretrained model (ner_opioid) from Spark NLP Healthcare to extract NERs from text and
filter them based on assertions and relations.
-Use the NERs to create a KG.
-Using RE models, NERs can be linked to each other.
-Example – which opioid caused which condition (even if there are multiple opioids in the same article).
24

25. 25

26. 26

27. Thank you!
27
Gursev Pirge
[email protected]
https://www.johnsnowlabs.com/