Skip to content

CONTRIBUTION.md

Latest commit

 

History

History
129 lines (90 loc) · 4.14 KB

CONTRIBUTION.md

File metadata and controls

129 lines (90 loc) · 4.14 KB

Contributing to Biomni

Thank you for your interest in contributing to Biomni! We're building the infrastructure layer for biomedical AI agents, and we welcome contributions from the community. Contributors with significant contributions will be invited to co-author publications in top-tier journals and conferences.

Getting Started

Before contributing, please ensure you:

  • Have tested your changes locally
  • Follow the existing code style and conventions
  • Include appropriate documentation

Types of Contributions

🛠️ Adding a New Tool

Tools are implemented as Python functions in biomni/tool/XXX.py, organized by subject area.

Steps:

  1. Implement and test your function locally. If it requires additional software, create installation script and append it into biomni_env/new_software_{VERSION}.sh

  2. Choose the appropriate subject category (e.g. database, biochemistry, etc.)

  3. Create a tool description in biomni/tool/tool_description/XXX.py following the existing format

    Tip: Use this helper to auto-generate descriptions:

    from biomni.utils import function_to_api_schema
from biomni.llm import get_llm

llm = get_llm('claude-sonnet-4-20250514')
desc = function_to_api_schema(function_code, llm)

  4. Create a test prompt that uses your tool and verify the agent works correctly

  5. Submit a pull request for review, don't forget to include your test prompt as well

📊 Adding New Data

If the data source has web API, follow this process:

Steps:

  1. Verify uniqueness - ensure no overlap with existing data
  2. Add a new query_XX function to biomni/tool/database.py, follow the format from the other functions.
  3. Create a tool description in biomni/tool/tool_description/database.py following the existing format

If the data source has no API access, follow the process below:

Steps:

  1. Verify uniqueness - ensure no overlap with existing data
  2. Prepare download link with verified redistribution rights
  3. Add entry to data_lake_dict in biomni/env_desc.py
  4. Submit a pull request with the download link

Then, make a PR.

💻 Adding New Software

Steps:

  1. Test locally to ensure no conflicts with existing environments
  2. Create installation script and append it into biomni_env/new_software_{VERSION}.sh
  3. Add entry to library_content_dict in biomni/env_desc.py
  4. Submit a pull request including:
    • Installation bash script
    • Screenshot demonstrating no environment conflicts

🎯 Adding a New Benchmark

Create benchmarks in the biomni/task/ folder.

Required implementation:

class YourBenchmark:
    def __init__(self):
        # Initialize benchmark
        pass

    def __len__(self):
        # Return dataset size
        pass

    def get_example(self, index):
        # Return dataset item at index
        pass

    def evaluate(self):
        # Evaluation logic (flexible input format)
        pass

    def output_class(self):
        # Return expected agent output format
        pass

Steps:

  1. Create benchmark file in biomni/task/[benchmark_name].py
  2. Implement required methods as shown above
  3. Provide data download link for associated datasets
  4. Submit a pull request

🐛 Bug Fixes & Enhancements

We welcome all bug fixes and enhancements to the existing codebase!

Create an issue to discuss with the Biomni team first.

Guidelines:

  • Clearly describe the issue or enhancement
  • Include tests when applicable
  • Follow existing code patterns
  • Update documentation if needed

Submission Process

  1. Fork the repository
  2. Create a feature branch from main
  3. Make your changes following the guidelines above
  4. Test thoroughly in your local environment
  5. Submit a pull request with a clear description

Review Process

The Biomni team will review all pull requests promptly. We may request changes or provide feedback to ensure code quality and consistency.

Questions?

If you have questions about contributing, please open an issue or reach out to the maintainers.

Together, let's build the future of biomedical AI agents!