Write up proposal

[MiguelRodo]

Tasks

• [ ] Describe basic use case
• [ ] Ask if it's worthwhile writing this up as a re-usable pipeline
• [ ]
• [ ] Describe how you transform characters to formulae
• [ ] Ask about whether you can use static branching

[MiguelRodo]

Describe basic use case

• Let's say you want to model various outcomes against some basic subset.
• y1 ~ x, y2 ~ x, y3 ~ x and so on.
  • Perhaps the responses are gene expression for various transcripts, or frequencies of different cell subsets.
  • This kind of analysis is done all the time.
• Then let's say that you want to fit a specifical model to that. Let us be simple and say we just pass the function, e.g. glmmTMB::glmmTMB.
• Then you fit that model to all these different formulae, and save the resultant models and extract the p-values.
• The results are catalogued in a dataframe, where each row corresponds to a model and there is a column for the dependent variable and another for the explanatory variable.
• I'd like to be able to write this up as a re-usable targets pipeline, with the following features:
  • Flexible formulae specification
  • My idea is to have the user specify the model by providing character vectors (for standard coefficients) and lists (for more complicated variables) to the variables dep (for dependent variable),exp(for explanatory variable, re (for random effects) and offset (for offset variables).
  • For example:
    • dep = "y1" means that the dependent variable is in column y1 of the dataframe.
    • exp = "x1" means that the explanatory variable is in column x2 of the dataframe.
    • exp = c("x1", "x2") means that the explanatory variables are x1 + x2.
    • exp = list("cn_orig" = "x1", trans = log) means that the explanatory variable is x1 and must be logged.
    • exp = list("cn_orig" = "x1", "cn_new" = "x1_log", trans = log) means that the explanatory variable is based on x1 that is logged and saved as a new variable x1_log.
  • There's more to it, but that's kind of the idea.
  • User-specifiable models
  • I know this is not "user-friendly", but my initial idea is that a model is specified as a list with values "pkg" for the R package, "fn" for the specific function and "args" for additional, non-formulae or data arguments.
• You have many models to fit to different responses or differe