scGPT: end-to-end protocol for fine-tuned retinal cell type annotation.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution cellular analysis but faces challenges in accurate cell-type annotation, particularly with large datasets and rare cell types.
• Foundation models, such as the transformer-based single-cell generative pretrained transformer (scGPT), offer scalable and flexible solutions for complex biological data analysis.

Purpose of the Study:

• To provide a comprehensive protocol for fine-tuning scGPT for cell-type classification in scRNA-seq data.
• To demonstrate the application of scGPT on a custom retina dataset for improved annotation accuracy and efficiency.
• To offer an accessible workflow for researchers to deploy scGPT for their specific datasets.

Main Methods:

• Fine-tuning of the scGPT foundation model using a custom retina scRNA-seq dataset.
• Development of an automated protocol encompassing data preprocessing, model fine-tuning, and evaluation.
• Provision of user-friendly tools, including a command-line script and Jupyter Notebook, for model customization and exploration.

Main Results:

• Achieved high-precision cell-type annotation with a 99.5% F1-score on the custom retina dataset.
• Demonstrated the efficiency of scGPT in handling complex single-cell data and improving annotation accuracy.
• Successfully automated key steps in the cell-type classification workflow.

Conclusions:

• The developed protocol enables efficient and accurate cell-type annotation using scGPT for scRNA-seq data.
• The accessible workflow and provided tools empower researchers, even those with limited programming experience, to leverage scGPT.
• This work offers an off-the-shelf solution for high-precision cell-type annotation, advancing single-cell data analysis.