Default: update.tree = TRUE

[TuomasBorman]

Agglomeration and subsetting functions have update.tree parameter that controls whether to simplify the the tree to correspond the data after agglomeration/subsetting. The default choice should be that the tree is updated, i.e., update.tree = TRUE.

[TuomasBorman]

Also update examples and OMA to reflect this new default choice

[antagomir]

How about the node labels?

[TuomasBorman]

Not sure if I answer to your question, but the names of tree nodes are not updated. The pruning is done internally by utilizing TreeSummarizedExperiment::subsetByLeaf(). It selects those nodes that are represent in nodes. Moreover, it simplifies the tree by removing those branches that have only one descendant.

mia/R/agglomerate.R

Line 494 in 2dc3430

x <- do.call(subsetByLeaf, args)

[antagomir]

What I meant was whether it would make sense to also update node labels so that the user could, by default, avoid cryptic errors like this:

library(mia)
data(GlobalPatterns)
tse <- GlobalPatterns

# Agglomerate by Species and subset by prevalence
tse <- agglomerateByRank(tse, rank="Family", update.tree = TRUE)

# Run MDS with UniFrac
library(scater)
tse <- runMDS(tse,
              FUN = getDissimilarity,
              name = "Unifrac",
              method = "unifrac",
              tree = rowTree(tse),
              ntop = nrow(tse),
              assay.type = "counts",
              weighted = TRUE)

Error: Incompatible tree and abundance table! Please try to provide 'node.label'.

[TuomasBorman]

Ahh, yes. It should be relatively easy to do that. And I think this is good idea.

Related to runMDS, we could have addMDS and getMDS that have default value for the dissimilarity function. That would make the usage much simpler.

[antagomir]

Sounds like a good idea!

[RiboRings]

Just as a note, the cryptic error doesn't go away when using update.tree = FALSE or update.tree = TRUE alike:

library(mia)
library(scater)

data(GlobalPatterns)
tse <- GlobalPatterns

# update.tree = TRUE

tse2 <- agglomerateByRank(tse, rank = "Family", update.tree = TRUE)
 
tse2 <- runMDS(tse2,
               FUN = getDissimilarity,
               name = "Unifrac",
               method = "unifrac",
               tree = rowTree(tse2),
               ntop = nrow(tse2),
               assay.type = "counts",
               weighted = TRUE)

Error: Incompatible tree and abundance table! Please try to provide 'node.label'.

# update.tree = FALSE

tse2 <- agglomerateByRank(tse, rank = "Family", update.tree = FALSE)

tse2 <- runMDS(tse2,
               FUN = getDissimilarity,
               name = "Unifrac",
               method = "unifrac",
               tree = rowTree(tse2),
               ntop = nrow(tse2),
               assay.type = "counts",
               weighted = TRUE)

Error: Incompatible tree and abundance table! Please try to provide 'node.label'.

[TuomasBorman]

Yep, that is because the names of nodes are not updated even though the tree is pruned

[antagomir]

And based on the above discussion the consensus seems to be to change this so that also node labels would be updated by default.