A reference-free approach for cell type classification with scRNA-seq.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution cellular characterization.
• scRNA-seq data suffers from sparsity due to low sequencing depth and read discarding.
• Existing methods often rely on gene expression, which can be suboptimal with sparse data.

Purpose of the Study:

• To develop a novel, reference-free, and alignment-free method for cell type classification using scRNA-seq data.
• To address the challenge of data sparsity in scRNA-seq by utilizing raw reads.
• To improve the accuracy of cell type classification compared to traditional gene expression-based methods.

Main Methods:

• Proposed scSimClassify, a method utilizing k-mer level features.
• Introduced compressed k-mer groups (CKGs) identified via simhash for feature extraction.
• CKGs capture k-mers with similar abundance profiles, serving as cell features.

Main Results:

• CKG features demonstrated superior performance in scRNA-seq classification accuracy compared to gene expression features in most cases.
• The reference-free and alignment-free nature of scSimClassify allows for effective utilization of raw reads.
• Achieved improved classification accuracy, particularly beneficial when reference genomes are incomplete or insufficient.

Conclusions:

• scSimClassify offers an effective alternative for cell type classification in scRNA-seq.
• The CKG feature representation overcomes data sparsity and enhances classification accuracy.
• This approach is particularly valuable for studies involving non-model organisms or incomplete genomic references.