A rank-based marker selection method for high throughput scRNA-seq data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-resolution data from millions of cells, enabling studies of rare cell types and developmental pathways.
• Identifying specific cell populations requires selecting informative genetic markers from scRNA-seq count data.
• Existing marker selection tools often rely on complex statistics and handle multi-class problems heuristically.

Purpose of the Study:

• To introduce RANKCORR, a computationally efficient and mathematically robust method for multi-class marker selection in scRNA-seq data.
• To develop novel performance metrics for evaluating marker set quality without a known ground truth.
• To compare RANKCORR's performance against existing methods across diverse datasets.

Main Methods:

• RANKCORR utilizes a non-parametric approach by ranking mRNA count data.
• The method performs linear separation of ranked data using a minimal set of genes.
• Performance metrics were developed to assess marker set quality in the absence of ground truth.

Main Results:

• RANKCORR demonstrates strong performance, consistently ranking among the most optimal marker selection methods.
• The method effectively handles multi-class marker selection in scRNA-seq data.
• Comparisons on experimental and synthetic datasets show competitive results, especially for large-scale data.

Conclusions:

• RANKCORR is a highly efficient tool for marker selection in high-throughput scRNA-seq data.
• Its speed makes it suitable for analyzing the largest datasets, outperforming slower methods.
• The software is available with extensive documentation, facilitating its integration into computational pipelines.