Trajectory-guided dimensionality reduction for multi-sample single-cell RNA-seq data reveals biologically relevant...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Analyzing multi-sample single-cell RNA sequencing (scRNA-seq) data poses significant dimensionality reduction challenges.
• Existing methods struggle to capture complex biological variations across different samples and conditions.

Purpose of the Study:

• To introduce MUlti-Sample Trajectory-Assisted Reduction of Dimensions (MUSTARD), a novel dimensionality reduction technique for multi-sample scRNA-seq data.
• To enable simultaneous analysis of gene expression variation along pseudotime and across samples.

Main Methods:

• MUSTARD integrates pseudotemporal information for an unsupervised approach.
• It captures major gene expression patterns along trajectories and across multiple samples.

Main Results:

• MUSTARD demonstrated superior performance in distinguishing sample groups and out-of-sample prediction accuracy in simulations.
• The method identified biologically relevant variations in COVID-19 and tuberculosis datasets, including links to symptom severity and immune responses.
• Significant overlap in immune response genes was observed across independent COVID-19 datasets.

Conclusions:

• MUSTARD is a flexible and powerful tool for uncovering biologically meaningful sample heterogeneity in diverse scRNA-seq datasets.
• The approach facilitates the discovery of endotypes and associated gene markers.