A comparison of marker gene selection methods for single-cell RNA sequencing data

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) allows detailed analysis of cellular transcriptional heterogeneity.
• Marker gene selection is a crucial step in scRNA-seq data analysis for cell type annotation.
• Numerous computational methods exist for marker gene identification, necessitating a comprehensive evaluation.

Purpose of the Study:

• To benchmark and compare the performance of 59 computational methods for marker gene selection in scRNA-seq data.
• To assess methods based on their accuracy in identifying known marker genes, predictive performance, and computational efficiency.
• To provide guidance on selecting appropriate marker gene identification tools for scRNA-seq analysis.

Main Methods:

• Evaluation of 59 marker gene selection algorithms using 14 real and over 170 simulated scRNA-seq datasets.
• Comparison criteria included recovery of known marker genes, gene set characteristics, speed, memory usage, and implementation quality.
• Case studies were performed to scrutinize commonly used methods and identify potential issues.

Main Results:

• Performance varied significantly across the 59 evaluated methods.
• Simple statistical tests, including the Wilcoxon rank-sum test, Student's t-test, and logistic regression, demonstrated high efficacy.
• The study identified inconsistencies and issues with certain commonly used marker gene selection approaches.

Conclusions:

• Simple methods, particularly the Wilcoxon rank-sum test, Student's t-test, and logistic regression, are effective for marker gene selection in scRNA-seq data.
• A comprehensive benchmark provides valuable insights for researchers choosing methods for cell type annotation.
• The findings emphasize the utility of established statistical approaches in modern single-cell genomics analysis.