From b2d463fdb7ffe71acc470601febdfbc02d909d9c Mon Sep 17 00:00:00 2001 From: Alan O'Callaghan Date: Sun, 28 Jan 2024 22:27:05 +0000 Subject: [PATCH] Update with jan 24 comments --- Workflow.Rmd | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/Workflow.Rmd b/Workflow.Rmd index 19aaa0b..9dee167 100644 --- a/Workflow.Rmd +++ b/Workflow.Rmd @@ -624,7 +624,10 @@ The output from `BASiCS_MCMC` is a `BASiCS_Chain` object that contains the draws associated to all model parameters. Given that `(N - Burn) / Thin = (30,000 - 15,000) / 15 = 1000` the object contains 1,000 draws for each parameter. -These can be accessed using the `displayChainBASiCS` function. +The matrices of MCMC draws can be accessed using the `displayChainBASiCS` +function --- this +may be useful for estimating and visualising credible intervals using +packages like `r CRANpkg("bayesplot")` or `r CRANpkg("tidybayes")`. For example, the following code displays the first 2 MCMC draws for mean expression parameters $\mu_i$ associated to the first 3 genes. @@ -1654,9 +1657,12 @@ posterior summaries provided by Stan [@Carpenter2017], while also being fully compatible with the workflow described in this manuscript via the function `Stan2BASiCS`, that converts the output of the Stan inference procedure to the type of output generated by `BASiCS_MCMC`. However, we note that -the inference methods provided by Stan are often very slow (Hamiltonian -Monte Carlo) or unreliable (variational inference, maximum *a posteriori*) -for BASiCS. +the Hamiltonian Monte Carlo inference method provided by Stan is +more computationally intensive than our default implementation. +Furthermore, the facster approximations provided by Stan, namely +scalable variational inference and maximum *a posteriori* estimation, +are often unstable and less accurate. + Finally, we also anticipate potential extensions of `r Biocpkg("BASiCS")` to account for the more complex experimental designs. For example, in cohort @@ -1702,6 +1708,7 @@ on EBI ArrayExpress under accession number The MCMC chains used to generate this article can be found in Zenodo under the DOI [10.5281/zenodo.10251224](https://doi.org/10.5281/zenodo.10251224). + # Software availability **Software: All software used in this workflow is available as part of