Better documentation

[gabrielodom]

Despite the volume of documented functions, the documentation itself is Don't RY in places where we need repetition and Do RY in places we don't. I need a fresh pair of eyes to look at this package documentation and tell me what places don't have the right documentation, and what places have superfluous or extraneous documentation.

[gabrielodom]

One example: we know that the CoMethAllRegions() function calculates leave-one-out probe correlations, but this isn't actually stated anywhere except in CreateRdrop() (which is like 4-5 function calls deep). Even worse: we don't mention how the correlations between a single probe and all other probes in the region are found (we take the row mean for the other probes, but that's not mentioned anywhere).

[gabrielodom]

Also in CoMethAllRegions(), the the default value for output is CpGs (which is correct), but maybe we need to write a function that will extract the CpGs just in case the user selected dataframe. A simple map() call won't work, because it ignores which CpGs pass the rDrop threshold and should be kept.

[DarioS]

Could you add some discussion of tissue analysis versus cell line analysis to the F.A.Q.? For example, if a data set has a sample that is 80% cancer cells and another sample that is 20% cancer cells, correlations may be induced simply by differences of cell type proportions from one sample to the next, rather than the addition or removal of a methyl group from a cytosine (what most people are looking for but often don't realise that the data which they have generated does not suit their question).

[gabrielodom]

@lxw391 do you have thoughts on this? Can we discuss it?

[lxw391]

@DarioS @gabrielodom It's a good suggestion. coMethDMR is flexible and the linear models in coMethDMR can include additional covariate variables to adjust differences in cell type proportions. For blood sample analysis, we can include estimated proportions of different types of cells (via R package such as EpiDish) as covariate variables in the linear model. If the cell type proportions can not be estimated, alternatively you can also use the sva R package to estimate surrogate variables that captures cellular composition. Hope this helps!