scACAN: An Adaptive Learning Framework Aggregating Local Graph Structure Context for Rare Cell Type Identification.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity.
• Existing methods struggle with uneven cell distribution and identifying rare cell populations.
• Adaptable models integrating contextual information are needed for scRNA-seq data.

Purpose of the Study:

• To introduce scACAN, an adaptive graph construction framework.
• To enhance the identification of both major and rare cell types in scRNA-seq data.
• To provide a robust and generalizable solution for single-cell data analysis.

Main Methods:

• scACAN utilizes aggregated local graph context information for positive sample selection.
• The framework incorporates adaptive sampling and iterative optimization based on clustering.
• scACAN is evaluated on multiple real-world scRNA-seq datasets.

Main Results:

• scACAN demonstrates superior performance in cell type identification.
• The method effectively identifies biologically significant rare cell subpopulations.
• Experiments confirm the robustness and generalizability of scACAN.

Conclusions:

• scACAN overcomes limitations in scRNA-seq analysis, particularly for rare cell types.
• The framework offers an effective approach for dissecting cellular heterogeneity.
• scACAN provides a valuable tool for advancing single-cell data analysis.