AnnoGCD: a generalized category discovery framework for automatic cell type annotation.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate cell type identification is crucial for understanding biological systems using single-cell RNA sequencing (scRNA-seq).
• Traditional methods struggle with the need for large, labeled datasets, which are often impractical due to cost and incomplete information.
• Discovering novel cell types remains a significant challenge in scRNA-seq data analysis.

Purpose of the Study:

• To develop a novel computational framework for automatic cell type annotation in scRNA-seq data.
• To address limitations of traditional methods by incorporating both labeled and unlabeled data.
• To enable the discovery of both known and novel cell types, including in imbalanced datasets.

Main Methods:

• Proposed AnnoGCD, a semi-supervised framework combining Generalized Category Discovery (GCD) and Anomaly Detection (AD).
• Implemented a two-block approach: a semi-supervised block for known cell type classification and an unsupervised block for novel cell type identification and clustering.
• Evaluated on five human scRNA-seq datasets and a mouse atlas.

Main Results:

• AnnoGCD demonstrated superior performance in identifying both known and novel cell types compared to existing methods.
• The framework showed robustness in handling datasets with significant class imbalance.
• Achieved accurate classification of known cell types and effective discovery of novel ones.

Conclusions:

• AnnoGCD offers a powerful and scalable solution for automatic cell type annotation in scRNA-seq data.
• The method advances biological research by enabling more comprehensive analysis of complex cellular populations.
• Provides a valuable tool for both biological research and clinical applications, with code available on GitHub.