Building a Pharmacogenomics Knowledge Model Toward Precision Medicine: Case Study in Melanoma

Area of Science:

• Pharmacogenomics
• Computational Biology
• Medical Informatics

Background:

• Individual genetic variations significantly impact drug efficacy and toxicity.
• Pharmacogenomics (PGx) is crucial for personalized medicine, yet lacks detailed drug usage data.
• Existing research has not fully explored the intricate relationships between PGx entities.

Purpose of the Study:

• To develop a pharmacogenomics knowledge model for uncovering hidden connections between drugs, genes, and diseases.
• To specifically detail precise medication parameters influenced by genetic factors.
• To enhance understanding of PGx for improved drug therapy.

Main Methods:

• Integration of open PGx data (DrugBank, RxNorm) and FDA drug labels.
• Manual annotation of 190 drug labels for entities and relationships.
• Training and evaluation of natural language processing models for entity recognition.

Main Results:

• A Bidirectional Encoder Representations from Transformers-conditional random field model achieved 85.12% micro-F1 score.
• The pharmacogenomics knowledge model encompasses 5 semantic types and 26 relationships.
• A case study on melanoma and BRAF mutations generated 4846 triples, detailing 7 related drugs.

Conclusions:

• The pharmacogenomics knowledge model offers a scalable framework for personalized drug dosage adjustments.
• It supports clinicians, pharmacists, and researchers in optimizing drug development and therapy.
• Future work includes expanding the model with more antitumor drugs and automated relation extraction.