InstaPrism

InstaPrism implements a fixed-point algorithm to accelerate BayesPrism, a Bayesian approach to computational cell-type deconvolution of bulk gene expression data, while maintaining accuracy and providing precompiled reference sets for cancer types.

Key Features:

• Fixed-point algorithm: Replaces Gibbs sampling with a derandomized fixed-point algorithm to reduce computational expense.
• Speed and memory efficiency: Provides considerable speed and memory advantages compared to the original BayesPrism method.
• Accuracy equivalence: Produces results effectively equivalent to BayesPrism in accuracy and robustness against model misspecifications.
• Precompiled reference sets: Includes curated reference sets tailored for a variety of cancer types.

Scientific Applications:

• Cell-type deconvolution: Models the compositional heterogeneity of bulk gene expression data by estimating cell type proportions and expression profiles.
• Cancer transcriptomics: Streamlines deconvolution analysis across multiple cancer types using precompiled references.

Methodology:

InstaPrism re-implements BayesPrism in a derandomized framework by replacing Gibbs sampling with a fixed-point algorithm.