scGAD: a new task and end-to-end framework for generalized cell type annotation and discovery

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression heterogeneity analysis.
• Cell annotation is crucial for downstream scRNA-seq data analysis.
• Existing methods struggle with novel cell types and batch effects.

Purpose of the Study:

• Introduce a generalized cell type annotation and discovery task for scRNA-seq data.
• Address limitations of current methods in handling novel cell types and batch effects.
• Propose a novel algorithmic framework, scGAD, for this task.

Main Methods:

• Developed a comprehensive evaluation benchmark for generalized cell type annotation and discovery.
• Proposed scGAD, an end-to-end framework using mutual nearest neighbors for anchor pairing.
• Implemented soft anchor-based and prototype self-supervised learning for label transfer and knowledge aggregation.
• Utilized a bidirectional dual alignment mechanism between embedding and prediction spaces.

Main Results:

• scGAD demonstrated superior performance over state-of-the-art methods on simulation and real-world datasets.
• The method effectively handles batch effects and cell type shifts.
• Marker gene identification confirmed scGAD's capability in clustering novel cell types and their biological significance.

Conclusions:

• scGAD offers a novel and practical solution for generalized cell type annotation and discovery in scRNA-seq data.
• The proposed framework effectively integrates known and novel cell type information.
• This work pioneers a new task and an end-to-end solution for enhanced scRNA-seq data analysis.