Highly accurate reference and method selection for universal cross-data set cell type annotation with CAMUS.

Area of Science:

• Single-cell genomics
• Computational biology
• Bioinformatics

Background:

• Cell type annotation is crucial for interpreting single-cell data.
• Existing reference-based methods offer rapid annotation but often lack optimal reference and method selection.
• This oversight can lead to suboptimal or inaccurate cell type assignments.

Purpose of the Study:

• To introduce a cross-dataset cell type annotation methodology with a universal reference data and method selection strategy (CAMUS).
• To enhance the accuracy and efficiency of cell type annotation in single-cell analyses.
• To provide a reliable strategy for selecting the best reference-method pairs.

Main Methods:

• Developed CAMUS, a novel methodology for selecting optimal references and methods for cell type annotation.
• Conducted comprehensive analyses on 672 pairs of cross-species single-cell RNA sequencing (scRNA-seq) datasets.
• Evaluated CAMUS performance across various single-cell data types including scRNA-seq, spatial transcriptomics (scST), and single-cell ATAC sequencing (scATAC-seq).

Main Results:

• CAMUS achieved substantial accuracy gains (25.0%-124.7%) compared to random selection strategies across five reference-based methods.
• CAMUS demonstrated high accuracy (49.1%) in selecting the optimal reference-method pair from 3360 possibilities.
• CAMUS showed high accuracy in selecting the best methods for scST (82.5%) and scATAC-seq (100.0%) data, indicating universal applicability.

Conclusions:

• CAMUS provides a robust and universally applicable strategy for accurate cell type annotation in single-cell data.
• The CAMUS score and associated metrics offer valuable guidance for assessing annotation reliability.
• This methodology addresses the critical need for optimized reference and method selection in single-cell analysis.