Some items to think about for the next steps (2023-10-28)

Target format

Do we want to provide the completed material as:
- an academic article with a detailed appendix?
- a persistent online guide? (e.g. bookdown, quarto html output?)
- both? (This is probably not much extra work, since an online guide would basically be an html rendering of a possibly slightly modified version of the material submitted to a journal.)
Ultimately, it would be nice to be able to produce something as helpful as Wilson 2010 "An ecologist's guide to the animal model".

Set out the problem that virulence in bacteria is being tested on animal models more regularly, but care needs to be taken in the analysis of the data and the design of the experiments.
We have a real life dataset which emphasises that correctly specifying the random effects means that "significance" goes away.
We can do simulations to also show this.
We can then provide some recommendations for thinking about your level of replication and your method of analysis.

It would be nice to provide an as simple as possible (but still rigorous) mathematical description for each presented model. Ideally, the reader could skip those maths sections without losing the ability to follow the guide, but at least the equations would be here if the reader want to go further into understanding if the model matches their assumptions.
This sounds good to me, but my understanding may also be limited so this always worries me a little! But could easily put it in the SI.

Vignette showing how failing to take into account random effects can lead to misleading results.
Vignette demonstrating how to choose an experimental design to improve power to detect differences in virulence: for example, in an evolution experiment where we have several bacteria clones evolved in different conditions and there is a maximum number of larvae we can afford to use for logistic reasons, is it better to use a few clones per treatment with more larvae per clone or to use more clones but with less larvae per clone?
This second point sounds awesome.

Suggestion: separating vignettes and manuscript material into two folders:
- ./vignettes/: contains Rmd (quarto files now probably) files, each file focusing on one topic (real life example, power analysis, comparison of analysis methods, ...). Used for prototyping, and ultimately we can generate polished figures and tables from them to include in the manuscript.
- ./manuscript/: contains files that are used to build the manuscript. We would have to choose which format we want to use (bookdown? quarto? latex?). Yikes. I have limited experience writing in anything apart from Google docs or Word, but maybe it would be fun to try quarto.
- ./tutorials/: contains files which will be shared as appendices to the manuscript, with detailed walk-through of different methods for the readers. Each could be a vignette which was initiated as a prototype in ./vignettes/ and was copied here once it was polished enough.
- I think tutorials could eventually be made into a website super easily where the tutorials/vignettes live.

Maybe one summary table would be nice, with columns being the different methods (coxph, coxme, rstanarm, brms, stan/jags) and rows being features (random effects, custom covariance structures, result format, ease of testing hypotheses, ease of visualization, ...).
Censored data.
Difference between knowing the exact time of death versus interval of death (e.g. if we check larvae every 6 hours). Also depends on the criteria for death (e.g. no response to external stimulation? Melanization? Buoyancy problems for fish?).
Increasing power = decreasing uncertainty in virulence estimates, i.e. getting narrower posteriors for the virulence parameters (or for differences or ratios of parameters of interest). Variation in posteriors comes from measurement noise and biological variation. Origins of variation: variability between larvae, between clones, differences in OD of injected cultures, differences in injected volumes, variation in temperature/moisture in incubation chambers, etc. Some sources of variation can be minimized or controlled for (e.g. technical replicates), but biological variation is inherent and can actually be the parameter of interest in some cases (e.g. variance partitioning, heritability studies).