This research demonstrates how word embeddings from biomedical literature can predict drug-gene relationships with high accuracy, creating new pathways for pharmaceutical research.

• Successfully applied analogy tasks to identify drug targets, achieving up to 75.8% precision at top 1
• Leveraged BioConceptVec embeddings trained on 30 million PubMed abstracts
• Demonstrated that even simple vector arithmetic can reveal hidden biological relationships
• Outperformed previous methods in drug-target prediction

This approach offers significant value for pharmaceutical companies by accelerating target identification, potentially reducing drug discovery costs and timelines while enabling more targeted therapeutic development.

Predicting Drug-Gene Relations via Analogy Tasks with Word Embeddings