cafferychen777/mLLMCelltype

Cell type annotation for single-cell RNA-seq using multi-LLM consensus

What it solves

mLLMCelltype is designed to automate the process of cell type annotation in single-cell RNA sequencing (scRNA-seq) data. Traditionally, identifying cell types from gene expression data is a manual or reference-dependent process; this tool removes the need for pre-training or reference datasets by leveraging the knowledge embedded in large language models (LLMs).

How it works

The framework uses a consensus-based approach where multiple LLMs (such as GPT, Claude, Gemini, and Qwen) analyze the same marker gene lists. Instead of relying on a single model, it integrates predictions from several models to reduce bias and errors. The process includes:

  • Iterative Discussion: LLMs evaluate evidence and refine their annotations through multiple rounds of deliberation.
  • Uncertainty Quantification: The system calculates Consensus Proportion and Shannon Entropy to flag annotations that the models disagree on.
  • Integration: It works directly with common single-cell analysis platforms like Scanpy (Python) and Seurat (R).

Who it’s for

It is intended for bioinformaticians and researchers performing single-cell transcriptomics analysis who want a more accurate, automated, and reference-free way to annotate cell clusters.

Highlights

  • Multi-Model Consensus: Combines 10+ LLM providers to improve accuracy over single-model baselines.
  • Reference-Free: Annotates cell types without requiring external reference datasets.
  • Cross-Platform Support: Available as both R and Python packages with a browser-based web application.
  • Transparency: Records the complete reasoning process used by the LLMs for each annotation.