glmGamPoi

glmGamPoi fits Gamma-Poisson generalized linear models to single-cell RNA-seq count data to model sampling variability and support downstream analyses such as differential expression, principal component analysis (PCA), and factor analysis.

Key Features:

• Modeling Sampling Variability: Uses the Gamma-Poisson distribution to model the sampling variability inherent in single-cell RNA-seq count data, including frequent zero and small counts.
• Efficient Parameter Inference: Performs parameter inference for Gamma-Poisson GLMs on large-scale single-cell datasets (including datasets with millions of cells) without requiring all data to be loaded into RAM.
• Handling Sparse Data: Exploits data sparsity to provide a statistically robust alternative to logarithm-like transformations for count normalization and modeling.
• Support for Downstream Analyses: Provides fitted models and parameter estimates that enable downstream analyses such as differential expression, PCA, and factor analysis.

Scientific Applications:

• Differential Expression Analysis: Identifying genes that are differentially expressed across conditions or cell types using modeled count distributions.
• Principal Component Analysis (PCA): Reducing dimensionality of single-cell RNA-seq datasets while accounting for count-level variability.
• Factor Analysis: Investigating latent variables that influence observed gene expression by modeling count data with Gamma-Poisson GLMs.

Methodology:

Fits Gamma-Poisson generalized linear models to count data and performs parameter inference optimized for large, sparse single-cell RNA-seq datasets by exploiting sparsity and avoiding loading all data into RAM.