BioMedAgent
Nature publishes article on biomedical AI agent — developed by the Chinese Academy of Sciences' ICT.
BioMedAgent is a self-evolving multi-agent bioinformatics analysis framework developed by the Institute of Computing Technology (ICT), Chinese Academy of Sciences, and published in Nature Biomedical Engineering in March 2026. When a researcher inputs the analysis goal in natural language, multiple LLM-based agents collaborate to autonomously perform the entire process, from establishing a multi-stage plan and selecting tools to executing code and summarizing results. Just as you would tell ChatGPT, "Find the differentially expressed genes in this RNA-seq data," by providing BioMedAgent with hypotheses and data, it generates a complete analysis report, from omics analysis to machine learning modeling, pathological image segmentation, statistical testing, and visualization. Existing LLM-based analysis tools have a structure where a single model handles a single task, making them limited in complex bioinformatics pipelines (workflows that require connecting dozens of tools in sequence). BioMedAgent solves this problem with two key components: the Interactive Exploration (IE) algorithm and the Memory Retrieval (MR) algorithm. IE is an interactive exploration mechanism where the agent actively adjusts the exploration path while analyzing intermediate results. MR is a self-evolving engine that systematically records successful tool combinations, workflows, and executed code in a memory bank for reuse in new tasks. Thanks to these two algorithms, BioMedAgent selects more accurate and efficient analysis paths as it gains experience. In the BioMed-AQA benchmark, which consists of 273 biomedical analysis questions, it achieved a 77% success rate, significantly surpassing ChatGPT-4o's 47%, and also demonstrated consistent performance on external BixBench datasets, proving its versatility. From the perspective of a biomedical researcher, the use cases for BioMedAgent are extensive. In clinical genomics research, by inputting the raw RNA-seq data of a patient cohort and instructing it to "find biomarkers related to survival," it can construct and execute an automated pipeline, including preprocessing (cel2matrix), differential expression analysis, survival curve (Kaplan-Meier) visualization, and statistical testing (t-test, QQ plot). In the field of precision medicine, it can integrate multi-omics data to autonomously build machine learning models for patient stratification, and in pathology, it can also handle cell segmentation and quantitative analysis of tissue images. The key value is that biologists or clinicians with no programming experience can perform these multi-stage analyses simply by using natural language prompts.
💻 System Requirements
GPU not required (LLM inference is delegated to the OpenAI API). GPU acceleration is possible for some local tools, such as pathology image segmentation.
Base package ~500MB, 5-10GB including Docker tool images.
⚡ Installation
### 4-1. Quick Start
```bash
git clone https://github.com/BOBQWERA/BioMedAgent.git
cd BioMedAgent
conda create -n BioMedAgent python=3.10 && conda activate BioMedAgent
pip install -r requirements.txt
export OPENAI_API_KEY="your_openai_api_key"
redis-server &
python demo.py --task ml
```
### 4-2. Detailed Installation
```bash
# 1. Clone the repository
git clone https://github.com/BOBQWERA/BioMedAgent.git
cd BioMedAgent
# 2. Create a Conda environment (Python 3.10)
conda create -n BioMedAgent python=3.10
conda activate BioMedAgent
# 3. Install dependencies
python -m pip install -r requirements.txt
# 4. Set the OpenAI API key
export OPENAI_API_KEY="your_openai_api_key"
# 5. Start the Redis server (requires prior installation)
redis-server
# 6. Run the demo (supported tasks: ml, statistics, visualization, omics)
python demo.py --task ml # Machine learning analysis
python demo.py --task statistics # Statistical tests (t-test, QQ plot)
python demo.py --task visualization # Visualization (survival curve, violin plot)
# 7. Additional tools for omics analysis (requires Docker image)
# cel2matrix environment → refer to the tool/ folder
# survival_curve, t_test tools → download Docker image (refer to the README link)
```🧬 Bio Use Cases
Clinical Genomics Biomarker Discovery
Input RNA-seq raw data and clinical metadata, and use the "Explore survival-related biomarkers" prompt → automatically construct a pipeline for cel2matrix preprocessing, DEG analysis, Kaplan-Meier survival curves, and Cox regression. Reduces analysis time from several days to several hours compared to manual methods.
Multi-Omics Patient Stratification
Integrate genomics, transcriptomics, and proteomics data, and use the "Build a model to predict treatment response" instruction → automatically perform feature selection, cross-validation, and ensemble ML model training, and generate a performance report. Enables non-computational researchers to perform independent analyses in precision medicine research.
Quantitative Analysis of Pathology Images
From tissue section images, use the "Segment tumor area and quantify cell density" prompt → automatically call image segmentation tools, calculate area, cell count, and density statistics, and generate a visualized report.
📝 Update Notes
No update notes yet.
🧪 Related Code of Life
No related Code of Life posts yet.