bayNorm

bayNorm applies a Bayesian model to normalize and impute single-cell RNA sequencing (scRNA-seq) counts and recover realistic transcript distributions for downstream analyses.

Key Features:

• Bayesian Framework: Performs scaling and inference on scRNA-seq counts using a Bayesian approach to model uncertainty and variability.
• Binomial Likelihood for mRNA Capture: Uses a binomial model to represent the stochastic mRNA capture process in sequencing.
• Empirical Bayes Priors: Estimates prior distributions from expression values across cells via an empirical Bayes approach.
• Global Scaling Normalization: Integrates likelihood and priors to perform global scaling normalization of count data.
• Robust Imputation: Imputes missing observations to generate transcript distributions that align with single molecule FISH measurements.
• Improved Differential Expression Analysis: Enhances differential expression sensitivity and accuracy by using priors informed by dataset structure.
• Batch Effect Reduction: Reduces batch effects to facilitate more consistent comparisons across experimental batches or datasets.

Scientific Applications:

• Single-cell expression quantification: Recovers normalized and imputed gene expression measurements at the single-cell level for accurate quantification.
• Differential expression analysis: Improves detection of gene expression differences across conditions or cell types.
• Imputation validated by smFISH: Provides imputed transcript distributions that are consistent with single molecule FISH measurements.
• Batch correction and cross-dataset comparison: Facilitates comparison of scRNA-seq data across batches and datasets by reducing technical variation.
• Downstream analyses: Supports downstream tasks such as clustering, trajectory inference, and gene regulatory network modeling.
• Biological studies: Applicable to research in complex biological systems, disease modeling, and developmental biology requiring precise single-cell quantification.

Methodology:

BayNorm integrates a binomial likelihood model for mRNA capture with empirical Bayes priors estimated from expression values across cells within a Bayesian framework to perform global scaling normalization and infer/impute scRNA-seq counts.