ReCellTy: Domain-specific knowledge graph retrieval-augmented LLMs workflow for single-cell annotation

2025-04-24

bioRxiv

AI Summary

Current single-cell annotation methods suffer from low automation, misalignment with human cognitive reasoning, and suboptimal performance of general-purpose large language models (LLMs). Method: We propose a knowledge-driven, graph-augmented LLM framework. It constructs a cell-type–marker knowledge graph, integrates differential gene–guided graph-based retrieval (RAG), and performs multi-task fine-tuning of the LLM, augmented by a semantic similarity alignment optimization strategy. Contribution/Results: This work pioneers deep synergy between domain-specific knowledge graphs and LLMs, explicitly emulating human annotation cognition during inference. Evaluated across 11 tissue datasets, our method achieves up to a +0.21 improvement in human expert evaluation scores and a 6.1% gain in semantic consistency—significantly outperforming general-purpose LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

To enable precise and fully automated cell type annotation with large language models (LLMs), we developed a graph-structured feature–marker database to retrieve entities linked to differential genes for cell reconstruction. We further designed a multi-task workflow to optimize the annotation process. Compared to general-purpose LLMs, our method improves human evaluation scores by up to 0.21 and semantic similarity by 6.1% across 11 tissue types, while more closely aligning with the cognitive logic of manual annotation.

Problem

• Enabling precise automated cell type annotation using LLMs
• Developing a graph database for cell reconstruction markers
• Optimizing annotation workflow to match manual cognitive logic

Innovation

• Graph structured feature marker database for retrieval
• Multi-task workflow optimizes annotation process
• Improves human evaluation and semantic similarity scores