Variance-adjusted Mahalanobis (VAM): a fast and accurate method for cell-specific gene set scoring

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis faces challenges due to technical noise and excessive zero counts.
• Existing gene set testing methods, optimized for bulk RNA-seq, perform poorly on scRNA-seq data.
• A critical gap exists in methods supporting cell-level gene set inference for scRNA-seq.

Purpose of the Study:

• To develop a novel gene set testing method tailored for scRNA-seq data.
• To address the limitations of existing methods in handling noise, sparsity, and large sample sizes.
• To enable robust cell-specific pathway scoring and statistical inference.

Main Methods:

• Developed Variance-adjusted Mahalanobis (VAM), a gene set testing method integrating with the Seurat framework.
• VAM computes cell-specific pathway scores, transforming gene expression matrices into pathway-level matrices.
• Utilized a gamma approximation for accurate population and cell-level inference.

Main Results:

• VAM effectively accommodates technical noise, sparsity, and large sample sizes inherent in scRNA-seq data.
• The method generates cell-by-pathway matrices suitable for data visualization and enrichment analysis.
• VAM demonstrated superior classification accuracy and lower computational cost compared to existing single-sample methods on simulated and real data.

Conclusions:

• VAM provides a powerful and efficient solution for gene set testing in scRNA-seq data.
• The method enhances data interpretation and statistical power by aggregating expression data at the pathway level.
• VAM supports both population- and cell-level inference, advancing scRNA-seq data analysis capabilities.