Version History & Changelog

In mLLMCelltype: Cell Type Annotation Using Large Language Models

knitr::opts_chunk$set(
  echo = TRUE,
  message = FALSE,
  warning = FALSE,
  eval = FALSE
)

Version History & Changelog

This document tracks the development history of mLLMCelltype, including major releases, feature additions, bug fixes, and other significant changes.

Version 1.0.0 (2023-11-15)

Initial Release

• First public release of mLLMCelltype
• Core functionality for cell type annotation using LLMs
• Support for OpenAI (GPT-3.5, GPT-4) and Anthropic (Claude) models
• Basic consensus mechanism
• Integration with Seurat

Version 1.1.0 (2024-01-20)

Features

• Added support for Google's Gemini models
• Implemented structured deliberation process for controversial clusters
• Added uncertainty quantification with consensus proportion and Shannon entropy
• Improved caching system for API responses
• Added Python implementation

Bug Fixes

• Fixed issue with marker gene sorting
• Corrected handling of cluster indices
• Improved error messages for API failures

Documentation

• Added comprehensive README
• Created example notebooks
• Added function documentation

Version 1.2.0 (2024-03-10)

Features

• Added support for DeepSeek, Qwen, and Zhipu models
• Implemented hierarchical annotation capability
• Added visualization functions for uncertainty metrics
• Improved handling of rate limits and API errors
• Added batch processing for large datasets

Bug Fixes

• Fixed consensus calculation for edge cases
• Corrected handling of empty API responses
• Improved error handling for network issues

Documentation

• Added benchmarking results
• Updated examples with new models
• Expanded troubleshooting guide

Version 1.3.0 (2024-05-15)

Features

• Added support for Stepfun, MiniMax, and OpenRouter models
• Implemented custom prompt templates
• Added provider-specific parameter customization
• Improved performance with parallel processing
• Enhanced caching with persistent storage

Bug Fixes

• Fixed issue with discussion logs formatting
• Corrected handling of special characters in marker genes
• Improved robustness against API changes

Documentation

• Added case studies for different tissue types
• Created advanced usage guide
• Updated installation instructions for all dependencies

Version 1.4.0 (2024-07-01)

Features

• Added support for Grok models from X.AI
• Updated Claude model support to include Claude 3.7 Sonnet
• Updated Gemini model support to include Gemini 2.5 Pro
• Improved consensus mechanism with weighted voting
• Enhanced visualization capabilities
• Added support for spatial transcriptomics data

Bug Fixes

• Fixed consensus result printing for controversial clusters
• Corrected model mapping for Claude 3.7 Sonnet
• Fixed undefined variable issue in consensus validation
• Improved Seurat object integration

Documentation

• Created comprehensive documentation website with pkgdown
• Added new vignettes for specific use cases
• Updated examples with latest models

Version 1.4.1 (2024-07-15)

Bug Fixes

• Fixed issue with printing consensus results for controversial clusters
• Corrected model mapping in anthropic.py to properly use Claude 3.7 Sonnet
• Fixed undefined has_names variable in consensus validation
• Improved test_pbmc3k.R and added to .gitignore to prevent API key leakage

Documentation

• Updated model lists in documentation
• Clarified API key setup instructions
• Improved error messages for common issues

Upcoming Features

The following features are planned for future releases:

Version 1.5.0 (Planned)

• Integration with additional single-cell analysis frameworks
• Enhanced support for multi-modal data
• Improved performance for large datasets
• Advanced visualization tools
• Expanded benchmarking across diverse tissues and species

Version 2.0.0 (Planned)

• Complete redesign of the consensus mechanism
• Integration with reference-based methods
• Support for custom model fine-tuning
• Interactive web interface for annotation review
• Comprehensive benchmarking framework

Breaking Changes

This section documents breaking changes that may require updates to your code.

Version 1.2.0

• Changed the default value of top_gene_count from 5 to 10
• Modified the return structure of interactive_consensus_annotation() to include additional metadata
• Renamed uncertainty_score to shannon_entropy for clarity

Version 1.3.0

• Changed the API for custom model registration
• Modified the caching system to use a different file structure
• Updated the required R version to 4.0.0 or higher

Version 1.4.0

• Changed the default models used in examples to Claude 3.7 and Gemini 2.5
• Modified the return structure of annotate_cell_types() to include more metadata
• Updated the required package versions for several dependencies

Deprecation Notices

The following features are deprecated and will be removed in future versions:

• simple_consensus() function (since v1.2.0): Use interactive_consensus_annotation() instead
• basic_visualization() function (since v1.3.0): Use the new visualization functions instead
• Support for older model versions will be gradually phased out as providers retire them

Acknowledgments

We would like to thank all contributors who have helped improve mLLMCelltype:

• Core developers: Chen Yang, [List other core developers]
• Contributors: [List contributors]
• Users who reported issues and suggested improvements

How to Cite

If you use mLLMCelltype in your research, please cite:

Yang, C., et al. (2024). mLLMCelltype: Multi-LLM Consensus Architecture for Robust Cell Type Annotation. bioRxiv 2024.04.10.647852; doi: https://doi.org/10.1101/2024.04.10.647852

Feedback and Contributions

We welcome feedback and contributions to improve mLLMCelltype. Please see our Contributing Guide for details on how to contribute.

Next Steps

Now that you've reviewed the version history, you can:

• Return to the introduction to get started with mLLMCelltype
• Explore advanced features to learn about the latest capabilities
• Check the FAQ for answers to common questions

mLLMCelltype documentation built on May 11, 2026, 9:06 a.m.