iDESC

iDESC performs differential expression analysis between two groups of subjects using single-cell RNA sequencing (scRNA-seq) data by modeling subject-specific effects and gene-expression-dependent dropout events.

Key Features:

• Zero-Inflated Negative Binomial Mixed Model: Employs a zero-inflated negative binomial (ZINB) mixed model to jointly account for count overdispersion and excess zeros.
• Subject Effect as Random Effect: Incorporates subject effect as a random effect within the log-mean component of the negative binomial distribution.
• Dropout Modeling via Gene-Level Pooling: Models dropout events that depend on gene expression levels by pooling information across genes to estimate dropout prevalence.
• Cell Type-Specific DE Analysis: Enables cell type-specific differential expression analysis while controlling for subject-related and technical variation.
• Simulation-Based Performance Evaluation: Demonstrated control of type I error rates and higher statistical power compared to eleven existing DE analysis methods in simulation studies.
• Real-Data Consistency and Disease Relevance: Applications to real scRNA-seq datasets showed improved consistency between datasets and relevance to disease conditions.

Scientific Applications:

• Multi-Subject scRNA-seq Studies: Analysis of heterogeneous scRNA-seq data from multiple subjects to control for environmental and genetic background variability.
• Disease-Related Expression Studies: Identification of disease-relevant gene expression changes across conditions or treatments.
• Cell Type-Specific Investigations: Discovery of cell type-specific differential expression signatures while accounting for dropouts and subject effects.

Methodology:

Fits a zero-inflated negative binomial mixed model with subject effect as a random effect in the log-mean component, models dropouts by pooling information across genes, and evaluates performance via simulation benchmarking against eleven existing DE methods and application to real scRNA-seq datasets.