Kieran R Campbell

when colnames(L) get named in case they are NULL with colnames(Y)

the inference runs through one ELBO computation and throws an error (below) if rownames(L) and colnames(Y) don't match or are both NULL. ``` onstructing tensorflow graph Removing 0 genes with...

This is currently handled via setting lambda to 1. Should instead modify the likelihood so it's Pois(0.1) or similar

Initialise all parameters from PCA etc to save MCMC runs

Accept seurat as input

1

Accept seurat data types as input using the `Convert` function in seurat. Note that `Seurat` should only be a suggests rather than depends to avoid depending on too many libraries.

For paper acceptance / bioconductor submission, we should aim to have the following usability enhancements: 1. Allow multiple ways of passing the marker matrix - ie let the user pass...

Intelligent subsetting of data

3

Casual user may want to pass in an entire SCE but just the rho matrix corresponding to their marker genes. In theory we should be able to detect this and...

I originally wrote that if an SCE isn't provided and a user provides a count matrix then effectively it should be cell by gene, ie ```r t( counts( sce )...

Use sizeFactors from SingleCellExperiment

3

If `s` is `NULL` we can check `sizeFactors(sce)` before computing them ourselves, but should have warning as often people don't know what sits there

Delta currently initializes to [log(min_delta), 2], which will break if log(min_delta) >= 2. Plan to change to [log(min_delta), log(min_delta) + c] for some c (2?), but need to check log(min_delta)...