BASiCStan

BASiCStan provides an interface to infer the parameters of BASiCS using variational inference (ADVI), Markov chain Monte Carlo (NUTS), and maximum a posteriori (BFGS) inference engines in the Stan programming language.

Key Features:

• Alternative inference engines: Supports ADVI, NUTS, and BFGS methods as alternatives to the adaptive Metropolis within Gibbs sampling scheme used by BASiCS.
• Stan integration: Leverages the Stan programming language for parameter inference of the BASiCS Bayesian hierarchical model.
• Scalability for complex posteriors: Alternative inference methods may be preferable for particularly large data or posterior distributions with difficult geometries.

Scientific Applications:

• Single-cell RNA sequencing analysis: Enables Bayesian inference of BASiCS model parameters for single-cell transcriptomic data using computationally efficient alternatives to Gibbs sampling.

Methodology:

BASiCStan implements parameter inference for the BASiCS Bayesian hierarchical model through Stan's variational inference (ADVI), Markov chain Monte Carlo (NUTS), and maximum a posteriori (BFGS) engines.