scROSHI: robust supervised hierarchical identification of single cells.

Area of Science:

• Single-cell genomics
• Computational biology
• Bioinformatics

Background:

• Accurate cell type identification is crucial for single-cell analysis.
• Current machine learning approaches often require annotated training data, which is scarce in early research.
• This limitation can result in overfitting and reduced performance on new datasets.

Purpose of the Study:

• To introduce scROSHI, a novel method for cell type identification in single-cell studies.
• To develop a tool that does not rely on annotated training data.
• To leverage hierarchical cell type relationships for improved prediction accuracy.

Main Methods:

• scROSHI utilizes pre-existing cell type-specific gene lists.
• The method does not require a training dataset.
• It assigns cell identities by respecting hierarchical relationships, moving from general to specialized types.

Main Results:

• scROSHI demonstrates excellent prediction performance.
• In benchmarks using peripheral blood mononuclear cell (PBMC) data, scROSHI outperformed existing methods.
• Superior performance was observed particularly when training data was limited or experimental diversity was high.

Conclusions:

• scROSHI offers a robust alternative for cell type identification, especially in data-limited scenarios.
• The method's hierarchical approach enhances prediction accuracy.
• It addresses key challenges in single-cell data analysis, improving discoverability and reliability.