HiCat: a semi-supervised approach for cell type annotation.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Supervised cell type annotation methods struggle with novel cell types.
• Unsupervised methods face challenges with cluster purity and distinguishing unknown cell populations.
• A gap exists in robustly annotating both known and novel cell types simultaneously.

Purpose of the Study:

• To develop HiCat, a semi-supervised pipeline addressing limitations in current cell annotation techniques.
• To improve accuracy in identifying known cell types and enhance the discovery of novel cell types.
• To provide a robust, scalable, and transferable solution for single-cell RNA sequencing (scRNA-seq) data analysis.

Main Methods:

• HiCat integrates reference (labeled) and query (unlabeled) genomic data.
• The pipeline involves batch effect removal, dimensionality reduction, unsupervised clustering, feature merging, supervised classification, and inconsistency resolution.
• A structured six-step process refines cell type annotations.

Main Results:

• HiCat demonstrated superior performance in classifying known cell types and identifying novel cell types across 10 public datasets.
• The pipeline excelled in distinguishing multiple novel cell types in benchmark evaluations.
• A case study on the human lung molecular cell atlas validated HiCat's effectiveness.

Conclusions:

• HiCat offers a robust framework for scRNA-seq cell annotation, improving both classification accuracy and novel type identification.
• The method effectively addresses key challenges in automated cell annotation.
• HiCat provides a scalable and transferable solution for biomedical research.