slalom

slalom provides a scalable Bayesian factor analysis framework for dissecting sources of variation in single-cell RNA-seq datasets.

Key Features:

• Bayesian factor analysis: Implements a scalable Bayesian factor analysis framework to model latent factors that drive cell-to-cell transcriptomic heterogeneity.
• f-scLVM implementation: Implements the factorial single-cell latent variable model (f-scLVM) in R/C++.
• Curated gene set integration: Leverages curated gene set annotations, including KEGG pathway sets, to infer interpretable latent factors.
• Annotated and unannotated factors: Models both annotated biological processes and additional unannotated factors concurrently.
• Gene-set membership refinement: Refines gene-set membership and estimates the relevance of each factor to the observed data.
• Joint modeling of confounders: Jointly captures technical effects, confounders, and structured biological drivers within the latent factor framework.
• Decomposition of variability: Decomposes scRNA-seq variability into meaningful components to reveal latent subpopulations and biological signals.

Scientific Applications:

• Pathway activity inference: Enables inference of pathway activity from single-cell transcriptomes using annotated gene sets.
• Confounder modeling: Facilitates modeling and adjustment for technical effects and other confounders in scRNA-seq data.
• Identification of cellular states: Supports identification of biologically coherent cellular states and latent subpopulations obscured by unwanted variation.
• Variance decomposition: Allows decomposition of cell-to-cell transcriptomic heterogeneity into interpretable biological and technical factors.

Methodology:

Uses a scalable Bayesian factor analysis implemented as the factorial single-cell latent variable model (f-scLVM) in R/C++, leveraging curated gene set annotations (e.g., KEGG), modeling annotated and unannotated factors, refining gene-set membership, estimating factor relevance, and jointly capturing technical effects, confounders, and structured biological drivers.