sctransform

sctransform normalizes and variance-stabilizes UMI-based single-cell RNA-seq (scRNA-seq) count data using regularized negative binomial regression to separate technical variability such as sequencing depth from biological heterogeneity.

Key Features:

• Normalization and Variance Stabilization: Computes Pearson residuals from a regularized negative binomial model to normalize molecular counts and reduce the influence of technical factors such as sequencing depth.
• Regularized Negative Binomial Regression: Fits a generalized linear model that includes sequencing depth as a covariate and regularizes parameter estimates by pooling information across genes of similar abundance to mitigate overfitting.
• Preservation of Biological Heterogeneity: Reduces technical noise while maintaining biological variability between cells for downstream analyses.
• Improved Downstream Analyses: Removes the need for heuristic pseudocount addition or log-transformation and enhances variable gene selection, dimensional reduction, and differential expression analysis.
• Compatibility with UMI-based scRNA-seq Data: Applicable to UMI-based single-cell RNA-seq datasets.
• Integration with Seurat: Provides a computational interface for use within the Seurat single-cell analysis toolkit.

Scientific Applications:

• Cellular heterogeneity analysis: Facilitates detection and characterization of heterogeneous cell states and populations in scRNA-seq datasets.
• Gene expression profiling: Enables more accurate gene-level expression comparisons across cells by controlling for sequencing depth and technical variance.
• Differential expression analysis: Improves differential expression testing across cell populations by reducing technical confounders and stabilizing variance.

Methodology:

Applies regularized negative binomial regression with sequencing depth as a covariate, pools information across genes of similar abundance to stabilize parameter estimates, and computes Pearson residuals to yield normalized, variance-stabilized counts for downstream analysis.