ProjectR - Return "full" output for NMF algorithm

[adkinsrs]

Currently when running projectR by itself, we run the command with full=FALSE, which returns just the projection weights. However, there was a request from a lab to return the full output which returns projectionFit, a list containing projectionPatterns and Projection. The Projection object contains additional information from the procedure used to obtain the projectionPatterns. Projection is the full lmFit model from the package limma (https://bioconductor.org/packages/3.19/limma).

Ideally we need to return this with the projection results and store it in caching so that it does not have to be retrieved in the future

Because we perform PCA and binary algorithms using native Python code, this option will not be available for them. It will also not be available for "fixednmf" because that uses the SJD R package instead of projectR

[adkinsrs]

I think a secondary issue related to this is how to incorporate these results. Currently the plot displays will reflect the original plot configuration except switching gene expression for pattern weights. We cannot access projection results from the dataset curator, so it seems we would have to have some toggle on the display to switch to the extra projection results, like p-value. Maybe instead of displaying it, it's just as simple as allowing for downloads of the extra information (i.e. tables)

In addition, we can also run PCA in projectR and get those extra results too, meaning we deprecate the straightforward Python version

[adkinsrs]

@carlocolantuoni, it might be nice to get your input (comments, suggestions) for the two comments above. For perspective, this was a request by the Fertig Lab more or less for a demo using their datasets as a buy-in to using gEAR.

I guess one thing I would want to know is if it is possible to return the same extra output as projectR when "full=TRUE" is passed (similar to limma lmFit contents or some equivalent). I think it might be less confusing if the user could select any available algorithm and have it return the "full" results... we'd still need to figure this out for Binary as well. Granted the easier path may just be to a have this only operational for "nmf"

[carlocolantuoni]

hey @adkinsrs - i agree that it would be difficult to think of a way to incorporate a way for the user to see/use these extra results. Elana Fertig (who requested this, and who is the new head of IGS) is a long time collaborator of mine - i can discuss with her what she is looking for here

[carlocolantuoni]

i sent her a note

[adkinsrs]

Thanks @carlocolantuoni. For added context, Joshua and I met with one of her lab people, Sushma, who mentioned they use those results in their work.