scALGSL: Active Learning and Graph Structure Learning for Cell Type Annotation From Single-Cell RNA-Seq Data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals tissue heterogeneity.
• Accurate cell type annotation is essential for scRNA-seq data interpretation.
• Current annotation methods face challenges like limited labeled data, poor graph structures, and lack of cell state information.

Purpose of the Study:

• To develop an innovative framework, scALGSL, for accurate cell type annotation in scRNA-seq data.
• To address limitations of existing annotation methods, including label scarcity, suboptimal graph topology, and missing cell state information.

Main Methods:

• scALGSL integrates dynamic graph optimization with active learning.
• A graph-guided active learning mechanism selects valuable training samples.
• A learnable graph structure optimization module refines adjacency matrices.
• A cell state auxiliary pathway extracts functional features using pre-trained models.

Main Results:

• scALGSL achieved an average accuracy of 0.896 and F1 score of 0.771 on a cancer dataset.
• The framework demonstrated robustness in cross-platform tasks.
• Integration of cell state information significantly improved performance.
• Ablation studies confirmed the necessity of node selection and edge optimization modules.

Conclusions:

• scALGSL offers a scalable solution for precise cell annotation in scRNA-seq data.
• This framework facilitates tumor microenvironment analysis and precision medicine applications.
• The developed method effectively overcomes key challenges in current cell annotation techniques.