scClassify: sample size estimation and multiscale classification of cells using single and multiple reference.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Automated cell type identification is crucial for analyzing single-cell RNA sequencing (scRNA-seq) data.
• Existing methods face challenges with diverse datasets and complex cell type hierarchies.

Purpose of the Study:

• To develop a robust and scalable framework for automated cell type identification in scRNA-seq data.
• To improve classification accuracy by leveraging ensemble learning and hierarchical structures.
• To assess the required sample size for accurate cell type classification.

Main Methods:

• Developed scClassify, a multiscale classification framework using ensemble learning.
• Constructed cell type hierarchies from single or multiple annotated scRNA-seq datasets.
• Employed simulations and experimental datasets for validation.
• Evaluated performance across 114 diverse reference and testing data pairs.

Main Results:

• scClassify demonstrated superior performance compared to other supervised cell type classification methods.
• The framework accurately estimates sample size requirements for cell type classification.
• scClassify successfully identified previously unidentified cell subpopulations in the Tabula Muris dataset.
• The method shows scalability on large single-cell atlases.

Conclusions:

• scClassify offers a state-of-the-art solution for automated cell type identification in scRNA-seq data.
• The framework enhances accuracy and provides insights into sample size needs.
• scClassify has broad applicability in analyzing complex single-cell datasets.