CureCrafter
Multi-Agent System for Drug Repurposing using Biomedical Knowledge Graphs
Created on 8th May 2025
•
CureCrafter
Multi-Agent System for Drug Repurposing using Biomedical Knowledge Graphs
The problem CureCrafter solves
CureCrafter enables autonomous biomedical research by using AI agents to extract gene–drug–disease relationships from paper abstracts and generate drug repurposing hypotheses. It solves the challenge of manually scouring the literature for hidden therapeutic insights by automating:
Information extraction from unstructured biomedical text
Building of a structured mini knowledge graph
Recommending alternative uses for existing drugs
This significantly reduces the time required for hypothesis generation in drug discovery, especially in rare or under-researched diseases.
Challenges I ran into
One major challenge was reliably extracting biomedical triples (Gene, Drug, Disease) from abstracts using LLMs. LLM responses were inconsistent, so we structured the prompt and added validation to handle exceptions.
Another challenge was designing an interpretable and scalable graph-based approach for repurposing suggestions. NetworkX was instrumental here, and building a fast prototype with FastAPI helped decouple the agents cleanly.
Also, deploying two separate microservices (PaperMinerAgent and BioGraphAgent) on Replit required careful handling of ports and payload validation.