glmGamPoi: fitting Gamma-Poisson generalized linear models on single cell count data.

Area of Science:

• Computational biology
• Statistical genetics
• Bioinformatics

Background:

• The Gamma-Poisson distribution is crucial for modeling single-cell RNA sequencing (scRNA-seq) count variability.
• Current implementations struggle with large scRNA-seq datasets and underutilize the prevalence of zero counts.
• These limitations hinder the adoption of this statistically robust model, favoring less accurate methods.

Purpose of the Study:

• To develop a faster and more accurate R package for fitting the Gamma-Poisson distribution to scRNA-seq data.
• To address the computational challenges posed by the scale and sparsity of modern single-cell datasets.

Main Methods:

• Development of the R package 'glmGamPoi' for parameter inference of the Gamma-Poisson distribution.
• Implementation of efficient algorithms optimized for large-scale, sparse single-cell count matrices.
• Capability to process data stored on disk, reducing RAM requirements.

Main Results:

• The 'glmGamPoi' package provides significantly faster and more accurate fitting of the Gamma-Poisson distribution compared to existing methods.
• The software efficiently handles large single-cell datasets, including those with millions of cells.
• The package supports out-of-memory computation, making it scalable for massive datasets.

Conclusions:

• 'glmGamPoi' offers a computationally efficient and accurate solution for fitting the Gamma-Poisson distribution in scRNA-seq analysis.
• The package facilitates the broader adoption of this statistically sound model for robust downstream analyses like differential expression.
• This tool addresses key limitations of existing methods, improving the analysis of complex single-cell genomics data.