Single-cell gene set scoring with nearest neighbor graph smoothed data (gssnng)

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene set scoring is crucial for interpreting biological changes in experiments.
• Traditional methods struggle with the sparsity and noise of single-cell RNA sequencing data.
• Existing methods were developed for bulk expression profiling, not single-cell resolution.

Purpose of the Study:

• To develop a robust method for gene set scoring in single-cell transcriptomics.
• To address the challenges posed by data sparsity and technical noise in single-cell expression measures.
• To generate high-quality, per-cell gene set scores for improved visualization and statistical analysis.

Main Methods:

• Utilizing a nearest neighbor graph of cells for matrix smoothing.
• Applying a smoothing transformation to share gene expression information within transcriptomic neighborhoods.
• Implementing the `gssnng` software package compatible with Scanpy AnnData objects.

Main Results:

• Achieved high-quality gene set scores at the per-cell, per-group level.
• Demonstrated the effectiveness of transcriptomic neighborhood smoothing for single-cell data.
• Enabled more reliable statistical modeling incorporating clinical factors like age and gender.

Conclusions:

• Transcriptomic neighborhood smoothing significantly improves gene set scoring for single-cell data.
• The developed method facilitates better biological interpretation and statistical analysis of single-cell experiments.
• The `gssnng` software provides an accessible tool for researchers in the field.