DeepRare is an agentic system designed to address the pervasive challenge of rare disease diagnosis by leveraging large language models (LLMs). It targets clinicians and researchers by providing ranked diagnostic hypotheses for rare diseases, complete with traceable, evidence-based reasoning chains. The system aims to improve diagnostic accuracy and timeliness, overcoming the heterogeneity and low prevalence issues characteristic of rare conditions.

How It Works

DeepRare employs a modular, agentic architecture comprising a central host with long-term memory and specialized agent servers. These agents integrate over 40 domain-specific tools and web-scale medical knowledge sources to process heterogeneous clinical inputs. The LLM-powered system generates diagnostic hypotheses, linking analytical steps to verifiable medical evidence, ensuring transparency and adaptability. This approach facilitates complex diagnostic reasoning while maintaining traceability.

Quick Start & Requirements

• Installation: Clone the repository (git clone https://github.com/MAGIC-AI4Med/DeepRare.git), navigate to the directory, and install dependencies (pip install -r requirements.txt).
• Prerequisites: Java 21+, Python 3.8+, ChromeDriver (version must match Chrome browser), and potentially Exomiser (requires ~20GB download and setup).
• LLM API Keys: Required for OpenAI, Anthropic, Google Gemini, DeepSeek, or custom local LLM endpoints. Keys are configured via environment variables (e.g., OPENAI_API_KEY).
• Hardware: Minimum 16GB RAM (32GB recommended), 100GB+ free SSD storage. GPU is optional but recommended for faster inference.
• Links: Web application available at http://raredx.cn/doctor.

Highlighted Details

• Achieves an average Recall@1 score of 57.18% on HPO-based evaluations, significantly outperforming 15 other methods, including traditional bioinformatics tools and LLMs.
• Demonstrates superior performance in multi-modal input scenarios, reaching 70.60% Recall@1 compared to Exomiser's 53.20%.
• Manual verification by clinical experts shows 95.40% agreement with the system's reasoning chains.
• Integrates over 40 specialized analytical tools and up-to-date medical knowledge sources.

Maintenance & Community

The project is associated with a publication in Nature, indicating significant research backing. Specific community channels (e.g., Discord, Slack) or maintainer details beyond the paper authors are not explicitly mentioned in the README.

Licensing & Compatibility

The README does not specify the software license. This omission requires clarification regarding usage restrictions, particularly for commercial applications or integration into closed-source systems.

Limitations & Caveats

The setup process is complex, requiring the installation and configuration of multiple external tools like ChromeDriver and Exomiser, alongside LLM API key management. While GPUs are optional, performance may be significantly slower on CPU-only systems. The absence of a stated license is a critical adoption blocker.