scExtract: leveraging large language models for fully automated single-cell RNA-seq data annotation and...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables the study of cellular heterogeneity.
• Analyzing large, unannotated public scRNA-seq datasets presents significant computational challenges.
• Existing methods for data annotation and integration often struggle with diverse datasets.

Purpose of the Study:

• To develop an automated framework, scExtract, for comprehensive scRNA-seq data analysis.
• To leverage large language models (LLMs) for efficient data preprocessing, annotation, and integration.
• To improve batch correction and preserve biological diversity in integrated scRNA-seq datasets.

Main Methods:

• scExtract utilizes LLMs to extract information from scientific literature for guided data processing.
• Introduced scanorama-prior and cellhint-prior methods for enhanced batch correction using prior annotation information.
• Benchmarked scExtract against existing reference transfer methods to assess performance.

Main Results:

• scExtract demonstrated superior performance in benchmarks compared to current reference transfer techniques.
• The scanorama-prior and cellhint-prior methods effectively improved batch correction while maintaining biological variation.
• Successfully integrated 14 scRNA-seq datasets, creating a large-scale human skin cell atlas.

Conclusions:

• scExtract offers an automated and efficient solution for analyzing complex scRNA-seq data.
• The framework enhances the utility of public scRNA-seq datasets through improved annotation and integration.
• The developed human skin atlas provides a valuable resource for future biological research.