Bray-Curtis issue

[antagomir]

It was reported by @nuorenarra that the vegan "bray" dissimilarity default does not correspond to the standard version of the widely known Bray-Curtis dissimilarity.

TODO: let us consider how the defaults should be set in mia (overriding vegan defaults?). At least this is something to mention in the documentation and OMA.

[antagomir]

@nuorenarra could you briefly describe the issue here?

[nuorenarra]

The issue is in jaccard distance, not in Bray-Curtis.

vegan package distance(method="jaccard") automatically considers abundance weighted version of jaccard, which is very similar to traditional Bray-Curtis distance. The standard definition of jaccard distance is not abundance-weighted, but simply count of shared species / sum of all unique species present in a pair. Vegan package distance function allows calculation of this traditional jaccard with an additional term distance(method="jaccard", binary=T), but since this is not well known, many mistakes have been made when people think vegan's jaccard refers to the traditional jaccard by default, which it doesn't.

[TuomasBorman]

1. We do not have calculateDistance function. It was removed some time ago. (I think the function could be something to include)

2. Check if runRDA function (dbrda and significance test) is working with binary parameter. --> OK, does work

3. Check if runMDS is working with binary parameter --> OK, works

4. Add example on jaccard distance with binary parameter to OMA

[antagomir]

We could consider adding calculateDistance; or it could be optional to return that as part of runMDS output. But that is a separate issue. I suggest we open an issue about that if a specific need will arise.

OK I like the solution, let us clearly mention this issue in OMA, with an example.

No need to add in packages (mia, miaViz, scater..)

[antagomir]

@nuorenarra when time allows, could you make the suggested PR to the documentation in mia and OMA on this?