BASiCS

BASiCS applies a Bayesian hierarchical model to analyze single-cell RNA sequencing (scRNA-seq) data by estimating cell-specific normalization constants and decomposing technical and biological variability to detect variable genes and changes in cellular heterogeneity.

Key Features:

• Normalization and technical noise quantification: Estimates cell-specific normalization constants within the model and quantifies technical variability using spike-in genes to decompose total expression count variability into technical and biological components.
• Detection of variable genes: Computes tail posterior probabilities to identify highly or lowly variable genes that reflect biological variance contributions across cells.
• Beyond-mean differential analysis: Identifies genes exhibiting shifts in cell-to-cell heterogeneity without significant changes in mean expression levels.
• Bayesian probabilistic framework: Uses Bayesian inference and posterior probabilities from a hierarchical model to provide probabilistic assessments of gene expression variability.
• Validation and case studies: Control experiments and case studies (including mouse Embryonic Stem Cells) demonstrated enrichment of gene ontology categories among genes identified as highly or lowly variable using cross-validation.
• Implementation: Implemented in R.

Scientific Applications:

• Analysis of cell-to-cell heterogeneity: Characterizes cell-specific biological variability in scRNA-seq datasets, including supervised experimental contexts.
• Differential variability discovery: Detects genes with altered heterogeneity across conditions that are not captured by mean-based differential expression tests.
• Functional enrichment validation: Supports downstream gene ontology enrichment analyses to validate biological relevance of variable gene sets, as shown in mouse Embryonic Stem Cells.

Methodology:

Estimating normalization constants and quantifying technical variability using spike-in controls; decomposing expression count variability into technical and biological components; utilizing Bayesian inference to calculate posterior probabilities that indicate gene variability within cell populations; implemented in R.