Evaluation of cell type annotation reliability using a large language model-based identifier.

Area of Science:

• Genomics
• Computational Biology
• Cellular Biology

Background:

• Accurate cell type annotation is crucial for single-cell RNA sequencing (scRNA-seq) data analysis.
• Existing methods face challenges due to bias and limitations in training data, leading to errors and inefficiencies.
• Developing robust and reliable annotation tools is essential for advancing cellular research.

Purpose of the Study:

• To develop a novel tool, LICT (Large Language Model-based Identifier for Cell Types), for accurate and reliable cell type annotation in scRNA-seq data.
• To overcome the limitations of existing annotation methods by leveraging multi-model integration and a flexible user interaction approach.
• To provide an objective framework for assessing annotation reliability and interpreting complex cellular phenotypes.

Main Methods:

• LICT utilizes a multi-model integration strategy powered by Large Language Models.
• A "talk-to-machine" approach facilitates intuitive interaction and annotation refinement.
• The tool was validated across diverse scRNA-seq datasets to assess its performance.

Main Results:

• LICT demonstrated consistent alignment with expert annotations across various datasets.
• The tool effectively interprets multifaceted cell populations, aiding in the discovery of biological insights.
• Comparisons showed LICT outperforms existing tools in efficiency, consistency, accuracy, and reliability.

Conclusions:

• LICT is a powerful and generalizable tool for scRNA-seq analysis, independent of reference datasets.
• Its objective framework enhances reproducibility and ensures more reliable results in cellular research.
• LICT empowers researchers to focus on biological discovery by streamlining the cell annotation process.