ext-ZINBayes

ext-ZINBayes applies a Bayesian zero-inflated negative binomial latent-variable model with variational inference to identify differentially expressed genes (DEG) in single-cell RNA sequencing (scRNA-seq) data while accounting for zero-inflation, overdispersion, and confounding technical and biological variability.

Key Features:

• Extended Bayesian Zero-Inflated Negative Binomial Factorization: Employs a latent-variable modeling framework combined with variational inference to model zero-inflation and overdispersion in scRNA-seq gene expression data.
• Benchmarking Against Established Methods: Benchmarked against scVI, SCDE, MAST, and DEseq using two public datasets: house mouse cells (two types) and human peripheral blood mononuclear cells (PBMCs) divided into four types.
• Competitive Performance: Benchmark results indicate competitive identification of putative biomarkers relative to the compared DEG methods.

Scientific Applications:

• Disease Profiling: Identification of differentially expressed genes at single-cell resolution to characterize disease states.
• Treatment Development: Elucidation of gene expression differences between healthy and diseased cells to support potential therapeutic target discovery.
• Cell Population Identification: Detection of novel cell populations based on distinct single-cell gene expression profiles.

Methodology:

Bayesian latent-variable modeling using a zero-inflated negative binomial likelihood with variational inference to infer latent variables that account for technical noise, zero-inflation, overdispersion, and biological variability in scRNA-seq data.