Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data contains technical variations, like differing molecule counts per cell, that can obscure true biological differences.
• Existing normalization methods may struggle to distinguish technical noise from genuine biological heterogeneity in scRNA-seq datasets.

Purpose of the Study:

• To develop a robust modeling framework for normalizing and stabilizing molecular count data from scRNA-seq experiments.
• To provide a method that effectively removes technical influences while preserving biological variation.

Main Methods:

• Proposed a normalization framework using Pearson residuals from regularized negative binomial regression.
• Incorporated cellular sequencing depth as a covariate in a generalized linear model.
• Addressed potential overfitting by pooling information across genes with similar expression levels for stable parameter estimation.

Main Results:

• Demonstrated that the proposed method successfully removes technical variation from scRNA-seq data.
• Showcased improved performance in downstream analyses, including variable gene selection, dimensional reduction, and differential expression analysis.
• Validated the approach's ability to preserve biological heterogeneity.

Conclusions:

• The regularized negative binomial regression approach offers a superior alternative to traditional normalization methods for scRNA-seq data.
• This method eliminates the need for heuristic steps like pseudocount addition or log-transformation.
• The framework is broadly applicable to UMI-based scRNA-seq datasets and is available in the R package sctransform.