mfa

mfa models bifurcations in single-cell transcriptomics using a Bayesian hierarchical mixture of factor analyzers to infer bifurcation structures and identify genes driving cell-fate decisions.

Key Features:

• Fully probabilistic generative model: Implements a fully probabilistic generative model for inferring bifurcation structures from single-cell gene expression data.
• Bayesian hierarchical framework: Uses a Bayesian hierarchical mixture of factor analyzers with full Markov-Chain Monte Carlo (MCMC) sampling for statistical inference.
• Automatic gene identification: Employs a hierarchical prior that facilitates automatic identification of genes that drive bifurcation processes.
• Zero-inflation handling: Includes an Empirical-Bayes-like extension tailored to zero-inflated single-cell RNA-seq data and quantifies when such models are beneficial.
• Validation and comparison: Demonstrated on real and simulated single-cell gene expression datasets with comparative analyses against existing pseudotime methods, while relying on computationally intensive full MCMC sampling.

Scientific Applications:

• Cellular differentiation modeling: Modeling bifurcations in single-cell RNA-seq to explore cellular differentiation pathways.
• Driver gene discovery: Identifying genes that drive bifurcation processes and cell-fate decisions.
• Developmental biology: Applying bifurcation inference to studies of developmental processes.
• Cancer research and personalized medicine: Supporting analyses in cancer biology and applications relevant to personalized medicine.

Methodology:

Bayesian hierarchical mixture of factor analyzers with hierarchical priors; full Markov-Chain Monte Carlo (MCMC) sampling for inference; an Empirical-Bayes-like extension to address zero-inflation; validation on real and simulated single-cell gene expression data and comparisons to pseudotime methods.