OpenBioMed is a Python deep learning toolkit designed for AI-powered biomedicine research, offering flexible APIs for multi-modal data (molecules, proteins, text, single cells) and over 20 tools for applications like drug discovery and multi-modal challenges. It enables researchers to build custom workflows and LLM-driven agents for complex scientific tasks.

How It Works

OpenBioMed employs a unified data processing pipeline to standardize heterogeneous biomedical data. It provides access to over 20 pre-trained models, including specialized foundation models like PharmolixFM for molecular and protein interactions, BioMedGPT-R1 for multimodal biomedical reasoning, and LangCell for single-cell transcriptomics. The framework supports building collaborative workflows and LLM agents, facilitating complex research by connecting multiple tools and leveraging large language models for insights.

Quick Start & Requirements

• Installation: Requires conda for environment setup. Key commands involve creating a conda environment (conda create -n OpenBioMed python=3.9), activating it (conda activate OpenBioMed), and installing PyTorch (v1.13.1 with specific CUDA version), PyTorch Geometric libraries, and project requirements (pip install -r requirements.txt).
• Prerequisites: CUDA toolkit (v11.7 recommended), Python 3.9. Additional installations are needed for visualization (PyMOL, imageio) and AutoDockVina (meeko, pdb2pqr, vina).
• Resources: Setup involves significant dependency installation. Docker images are available for easier deployment.
• Documentation: Tutorials are available as Jupyter notebooks.

Highlighted Details

• Features PharmolixFM, a multimodal foundation model achieving competitive performance with AlphaFold3 on protein-molecule docking.
• Includes BioMedGPT-R1, a multimodal biomedical reasoning model performing on par with commercial models on biomedical QA benchmarks.
• Offers LangCell for zero-shot cell identity understanding in single-cell transcriptomics.
• Provides BioMedGPT-10B, a multimodal foundation model for biomedical QA and cross-modal molecule/protein tasks.

Maintenance & Community

The project is a collaboration between PharMolix Inc. and the Institute of AI Industry Research (AIR), Tsinghua University. For technical questions or suggestions, users are encouraged to open an issue. Commercial support or collaboration inquiries can be directed to opensource@pharmolix.com.

Licensing & Compatibility

The repository is licensed under the MIT License. However, the use of specific models like BioMedGPT-LM-7B and BioMedGPT-10B is subject to an Acceptable Use Policy, which prohibits generating content that violates laws or regulations.

Limitations & Caveats

The Acceptable Use Policy for certain models strictly prohibits their use for generating harmful or illegal content. The project emphasizes responsible and ethical use, and BioMedGPT models should not be used to provide services to the general public.