Help & Documentation

Using the Chatbot

  • •Ask questions about drug-disease relationships, drug effectiveness, and drug-target analysis.
  • •The chatbot uses a knowledge graph built from over 150,000 PubMed publications.
  • •You can ask questions in natural language, and the system will retrieve relevant information from the knowledge graph.
  • •Press Enter to send a message, or Shift+Enter for a new line.

Knowledge Graph

  • •UniD³ employs a dual-stage entity extraction strategy to build hierarchical knowledge graphs.
  • •The knowledge graph contains information extracted from PubMed publications using Llama3.3-70B.
  • •The generated knowledge graphs and vector database are stored in zenodo.
  • •You can access the datasets through HuggingFace or download directly from zenodo.

Three Core Tasks

  • •Drug-Disease Matching (DDM): Identify relationships between drugs and diseases.
  • •Drug Effectiveness Assessment (DEA): Evaluate the effectiveness of drugs for specific conditions.
  • •Drug-Target Analysis (DTA): Analyze interactions between drugs and their molecular targets.

Datasets

  • •DDM, DEA, and DTA datasets are available on HuggingFace.
  • •You can access them using pandas or the HuggingFace datasets library.
  • •All datasets achieved F1 scores exceeding 0.80 across all tasks.
  • •The DDM task achieved an expert validation score of 0.9005 F1.

Quick Links

Knowledge Graph (Zenodo)Datasets (HuggingFace)

Frequently Asked Questions

How accurate are the responses?

The UniD³-generated datasets achieved F1 scores exceeding 0.80 across all three tasks, with expert validation scores reaching 0.9005 F1 in the DDM task.

What data sources are used?

The system leverages over 150,000 drug-related publications from PubMed, processed using Llama3.3-70B with carefully designed prompts.

Can I download the datasets?

Yes, all datasets (DDM, DEA, DTA) are available on HuggingFace and can be accessed using pandas or the HuggingFace datasets library.