A machine learning-based method for automatically identifying novel cells in annotating single-cell RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for dissecting tissue heterogeneity.
• Accurate cell type annotation is a critical first step in scRNA-seq data analysis.
• Supervised methods offer advantages over unsupervised clustering but struggle with novel cell type identification.

Purpose of the Study:

• To develop an automated method for identifying novel cell types in scRNA-seq data.
• To address the limitation of existing supervised methods in detecting cell types absent in training data.
• To provide a robust solution for comprehensive cell type annotation.

Main Methods:

• Developed a method combining autoencoder and iterative feature selection.
• Trained an autoencoder on labeled data and calculated reconstruction errors on testing data.
• Iteratively selected features with bi-modal patterns and reclustered cells.

Main Results:

• The proposed method accurately identifies novel cell types not present in training datasets.
• Combined with a support vector machine, it offers a complete cell annotation solution.
• Demonstrated superior performance compared to existing methods across five real scRNA-seq datasets.

Conclusions:

• The developed method effectively identifies novel cell types from scRNA-seq data.
• CAMLU enhances the accuracy and completeness of cell type annotation.
• The R package CAMLU is publicly available for broader use.