Clinical Agent is an advanced framework leveraging Large Language Models (LLMs) to enhance the efficiency and effectiveness of clinical trials.

Congratulations! This paper has been accepted at the 2024 ACM-BCB Conference!

ClinicalAgent/
├── algo/
│   ├── agents/
│   │   ├── tools/
│   │   │   ├── drugbank/
│   │   │   ├── enrollment/
│   │   │   ├── hetionet/
│   │   │   ├── risk_model/
│   ├── main.ipynb
├── web/

• /algo: Contains the core codebase for ClinicalAgent.
• /web: A preliminary web tool for ClinicalAgent based on LLM.

Before starting, set your OpenAI API key by adding the following lines to your ~/.bashrc file:

export OPENAI_API_KEY="sk-xxxxxxxxx"
export NEXT_PUBLIC_OPENAI_API_KEY="sk-xxxxxxxxx"

Ensure you have Python 3.8.19 installed. The algo directory requires the following dependencies:

• pytorch==1.12.1
• torchvision==0.13.1
• torchaudio==0.12.1
• cudatoolkit==11.3
• transformers==4.39.3
• tokenizers==0.15.1
• openai==1.28.0

To set up the environment, run:

python3.8 -m venv venv
source venv/bin/activate
pip install torch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit==11.3 transformers==4.39.3 tokenizers==0.15.1 openai==1.28.0

Before running ClinicalAgent, follow the README instructions in the drugbank, enrollment, hetionet, and risk_model directories to generate the necessary data for the tools:

• drugbank
• enrollment
• hetionet
• risk_model

The primary entry point for running ClinicalAgent is the main.ipynb notebook. This notebook utilizes various agents located in the algo/agents directory.

In main.ipynb, you can switch between using agents and tools or just plain GPT for answering questions by calling solve_problem() or solve_problem_standard(), as shown below:

subproblem_solve, final_result_str = solve_problem(user_problem)

subproblem_solve, final_result_str = solve_problem_standard(user_problem)

If you use ClinicalAgent in your research, please cite the following paper:

@article{yue2024ct,
  title={CT-Agent: Clinical Trial Multi-Agent with Large Language Model-based Reasoning},
  author={Yue, Ling and Fu, Tianfan},
  journal={arXiv preprint arXiv:2404.14777},
  year={2024}
}

Feel free to reach out if you have any questions or need further assistance.