TxAgent is an AI agent designed for precision therapeutics, offering personalized treatment recommendations by reasoning across a vast universe of biomedical tools. It targets researchers and clinicians needing to analyze drug interactions, contraindications, and patient-specific treatment strategies, aiming to improve therapeutic decision-making and reduce adverse events.

How It Works

TxAgent employs multi-step reasoning and real-time knowledge retrieval from 211 specialized tools within the ToolUniverse. It analyzes drugs at molecular, pharmacokinetic, and clinical levels, identifies contraindications considering patient comorbidities and concurrent medications, and tailors treatments based on individual characteristics like genetics and disease progression. The agent synthesizes evidence from multiple sources, assesses drug-condition interactions, and refines recommendations through iterative tool execution and structured function calls.

Quick Start & Requirements

• Installation: pip install txagent and pip install tooluniverse.
• Prerequisites: An H100 GPU with >80GB memory is recommended. ToolUniverse requires an internet connection.
• Demo: Run python run_txagent_app.py for a Gradio demo.
• Pretrained Models: Available on HuggingFace, including TxAgent-T1-Llama-3.1-8B.
• Resources: Project page with demo cases: https://github.com/mims-harvard/TxAgent

Highlighted Details

• Outperforms leading LLMs and reasoning agents on five new benchmarks (DrugPC, BrandPC, GenericPC, TreatmentPC, DescriptionPC) covering 3,168 drug reasoning tasks.
• Achieves 92.1% accuracy in open-ended drug reasoning, surpassing GPT-4o by up to 25.8%.
• Demonstrates strong generalization across drug name variants and descriptions, with variance < 0.01.
• Integrates multi-step inference, real-time knowledge grounding, and tool-assisted decision-making for clinically aligned recommendations.

Maintenance & Community

The project is led by researchers from Harvard University. Contact emails for questions are provided for Shanghua Gao and Marinka Zitnik.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The README strongly recommends an H100 GPU with over 80GB of memory, indicating significant hardware requirements for optimal performance. The licensing status is not clearly defined, which may impact commercial adoption.