GenoTEX
GenoTEX is a benchmark dataset for evaluating LLM-based agents on automated gene expression data analysis to identify disease-associated genes. It provides expert-curated annotations, code, and results for gene-trait association problems, including dataset selection, preprocessing, and statistical analysis. It is intended for researchers and developers working on AI-driven biomedical data analysis.
✨ Key features
- 1,384 gene-trait association problems (132 unconditional, 1,252 conditional)
- 41.5 GB of input data from GEO and TCGA
- 237,907 lines of expert-curated analysis code
- Three evaluation tasks: dataset selection, preprocessing, statistical analysis
- Comprehensive gene features (average 18,530 per dataset)
- Includes domain knowledge and gene synonym mapping
🎯 Use cases
- Benchmarking LLM agents for automated gene expression analysis
- Developing AI tools for disease gene discovery
- Evaluating data preprocessing and statistical analysis pipelines
- Studying conditional gene-trait associations with covariates
📦 Installation
🧰 Requirements: Python environment with sufficient disk space (82 GB for full data). No API keys required.
To install GenoTEX, clone the repository and download the data from the provided cloud storage links. For detailed installation, refer to the README. Since the README does not provide explicit installation commands, the following is a general guide:
git clone https://github.com/Liu-Hy/GenoTEX.git
cd GenoTEX
# Download data from Google Drive or Baidu Cloud and place in appropriate directories
🚀 Usage
The README does not provide a minimal usage example. For usage, refer to the code in the code/ directory and the tools in tools/. A typical workflow involves loading the data, running preprocessing, and performing regression analysis as demonstrated in the notebooks.
⚠️ Good to know
The benchmark is large (82 GB total) and requires substantial disk space; data is distributed via cloud storage or data platforms.
❓ FAQ
What is the total size of the GenoTEX dataset?
The total data size is 82.0 GB, as stated in the README.
How is the data distributed?
Data is available via Google Drive and Baidu Cloud Disk, or as complete datasets on Kaggle and Hugging Face.
What are the three evaluation tasks?
The three tasks are dataset selection, data preprocessing, and statistical analysis.
What is the structure of the regression output?
Each output file contains a significant_genes dictionary with lists of gene symbols, coefficients, and absolute coefficients, plus a cv_performance section.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.