• Collect all signatures as a list object, to be possibly used also later in data viz tasks (would also make it easier to curate/expand them)
• Think of having an intern function e.g. "compute_signature_generics", for the signature computation where all things are just staying the same

When using the compute_gold_standards function, I often get this warning associated to a long list of missing signature genes:

In compute.RIR(RNA.tpm) :
  differenty named or missing signature genes

Is the method robust enough to missing genes?
Shall we consider a gene symbol remapping step to make it more robust?

snake_case for

• function names
• file names
• function parameters

Careful to

• intra-function calls

Probably worth having a verbose parameter - this would control the output on the console as the code runs

Related to this: consider message vs print

The individual weightings of the genes in the 18-gene signature have been published in a patent filing, which can be found by a Google search on this number: WO2016094377 (Claim 21 a+c and Claim 23).

I received this answer from the original authors, not sure if we should display the weightings in compute_Tcell_inflamed.R

Hello,

I have a problem with compute_scores_immune_response funtion
i would like to get 10 score of immune response ("CYT", "Roh_IS", "chemokines", "Davoli_IS", "IFNy", "Ayers_expIS", "Tcell_inflamed", "RIR", "TLS", "Ock_IS"). but this function does not give me immune response score of "RIR" and "Ock_IS".

It also returned extra columns named 'resF_up', 'resF_down', and 'resF'. what is this column represent?

Could you please clarify or fix this?

• simplify calls to special tags
• try to indent to 80 chars/line or similar

The models were learned using longer names for the immune cells, so now the current names do not match the names in predict_immune_response.

• emails
• orcids
• more detailed Description: entry
• removing remotes once possible

• check tools in use if there's any special requirement
• if required, ask the authors if their license is not so clear

Adaptation to manuscript version

Should ideally directly follow the vignette creation, and precede/be combined with unit tests

• immunedeconv not in remotes
• quantiseqr to be finalized
• quantiseqr to be used

Trained models object is quite large, not compatible with "mere" software package - still, it is possible and encouraged to have a data package associated

This would also give an excellent opportunity to document the creation of the trained_models object at the heart of this

Run roxygen to update namespace after changing some functions names

In our analysis, we are actually not the expression of these genes as gold standard. If we decide not to use it, the function should be removed together with its functionality within compute_gold_standards. If we keep it, we should optimize the function which now is always being computed with compute_gold_standards.

... this stays up till we nail them all down, would be a milestone to checking in for the final submission

probably solvable with aes_string instead of aes