MINGLE: a mutual information-based interpretable framework for automatic cell type annotation in single-cell...

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Single-cell chromatin accessibility sequencing (scCAS) is crucial for understanding epigenomic heterogeneity.
• Existing methods face challenges in accurately annotating diverse cell types, especially rare or novel ones.

Purpose of the Study:

• To introduce MINGLE, an interpretable framework for accurate cell type annotation of scCAS data.
• To develop a novel strategy for identifying previously undiscovered cell types within scCAS datasets.
• To enhance the biological insights derived from scCAS data analysis.

Main Methods:

• Developed MINGLE, a mutual information-based framework leveraging cellular similarities and topological structures.
• Implemented a convex hull-based strategy for novel cell type identification.
• Conducted extensive experiments to evaluate MINGLE's performance against existing methods.

Main Results:

• MINGLE demonstrates superior cell type annotation accuracy, particularly for rare and novel cell types.
• The framework provides valuable biological insights, outperforming current annotation tools.
• MINGLE exhibits robustness and versatility across cross-batch, cross-tissue, and cross-species datasets, handling data imbalance and size variations effectively.

Conclusions:

• MINGLE offers a versatile and accurate solution for complex cell type annotation tasks in scCAS data.
• The framework's ability to identify novel cell types opens new avenues for biological discovery.
• MINGLE's performance across diverse data scenarios underscores its broad applicability in epigenomic research.