scEMAIL: Universal and Source-free Annotation Method for scRNA-seq Data with Novel Cell-type Perception.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Current single-cell RNA sequencing (scRNA-seq) annotation tools rely on source data, posing privacy challenges.
• Existing methods struggle with novel cell type discovery and often require subjective thresholds.
• Feature alignment between source and target data is difficult when raw source data is unavailable.

Purpose of the Study:

• To develop a universal annotation framework for scRNA-seq data that automatically detects novel cell types.
• To enable cell type annotation without direct access to source data, addressing privacy concerns.
• To improve the robustness and accuracy of cell type identification in scRNA-seq datasets.

Main Methods:

• Proposed scEMAIL framework with a novel cell-type perception module.
• Expert ensemble system for measuring cell uncertainty from multiple aspects.
• Bimodality tests and adaptive thresholding (manifold mixup) for novel cell detection.
• Model adaptation using global neighborhood messages and local affinity regularizations to mitigate batch effects.

Main Results:

• scEMAIL accurately and robustly detects novel cell types in both simulated and real scRNA-seq data.
• The framework successfully annotates cell types without requiring access to source data.
• Demonstrated flexibility and maintained superiority when applied to single-cell ATAC-seq data.
• Mitigated misclassifications by leveraging reliable self-supervised neighbor information.

Conclusions:

• scEMAIL offers a privacy-preserving and automated solution for cell type annotation in scRNA-seq data.
• The framework effectively identifies novel cell types without subjective user intervention.
• scEMAIL shows promise for diverse single-cell omics data, including scATAC-seq.