GPTAnno: Ontology-tree-guided hierarchical cell type annotation based on GPT models for single-cell data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptomic data analysis requires accurate cell type annotation, which is often hindered by challenges in determining optimal clustering resolution and maintaining consistent labeling across studies.
• Existing methods struggle with the inherent ambiguity in cellular clustering granularity and the lack of standardized annotation protocols, impacting data interpretability and reproducibility.

Purpose of the Study:

• To introduce GPTAnno, an automated, ontology-tree-guided, and uncertainty-aware hierarchical cell type annotation method leveraging GPT models.
• To provide a standardized, ontology-aware, and reproducible annotation solution that automatically selects optimal clustering resolutions and quantifies annotation uncertainty.

Main Methods:

• GPTAnno directly processes gene expression matrices, integrating multi-resolution clustering with large language model reasoning guided by a cell ontology.
• The method incorporates automatic selection of optimal clustering resolutions based on annotation distance within the ontology tree.
• Annotation uncertainty is quantified to identify ambiguous clusters requiring expert review.

Main Results:

• Benchmarking across twelve large-scale datasets demonstrated GPTAnno's superior accuracy in cell type annotation compared to existing methods.
• GPTAnno showed effectiveness across diverse species, tissues, and disease contexts.
• The method successfully produced standardized, ontology-aware, and reproducible annotations with reduced human effort.

Conclusions:

• GPTAnno offers a robust and automated solution for cell type annotation in single-cell transcriptomics, addressing key challenges in clustering and labeling.
• The method's ability to integrate large language models with cell ontologies and uncertainty quantification enhances annotation accuracy and interpretability.
• GPTAnno streamlines the annotation process, improving reproducibility and significantly reducing the manual workload for researchers.