BayesPrismExt: Enhanced Chain Storage and CV Computation

I've developed an extended version of the BayesPrism package with practical improvements for handling MCMC chain storage and coefficient-of-variation calculations in bulk RNA-seq deconvolution workflows.

Modifications

BayesPrismExt adds three optional arguments to the main run.prism and run.prism.st functions:

save.chain

Controls chain output with options for "none" (default, fastest), "theta" (cell-type fractions only), or "all" (both theta and Z expression matrices). Chains are stored in compressed HDF5 format with automatic chunking.

h5.file

Specifies the HDF5 output path (defaults to gibbs_chain.h5).

compute.z.cv

Enables coefficient-of-variation computation for gene-level Z parameters (disabled by default to reduce memory overhead). The standard theta.cv remains always computed.

Use Case

This is useful when you need to:

• Inspect posterior distributions directly after deconvolution
• Perform custom convergence diagnostics on MCMC chains
• Compute gene-specific uncertainty metrics without re-running the entire deconvolution
• Handle large chain arrays efficiently using HDF5 compression

The HDF5 format handles large chain arrays efficiently and includes a simple read.h5.chain() helper for extraction by sample index or name.