search

microbiome / OMA

Orchestrating Microbiome Analysis
https://microbiome.github.io/OMA
86 stars 43 forks source link

Modify the NMF example to use new getNMF wrapper #597

Closed thpralas closed 14 hours ago

thpralas commented 1 month ago

After PR https://github.com/microbiome/mia/pull/616, this PR modifies the NMF example in OMA.

The output of new wrapper getNMF is briefly explained, and the wrapper is used to calculate the results shown in the example.

TuomasBorman commented 1 month ago

I think this can be merged. Loadings could also be in rowData but I think this we discussed already.

This same thing applies to other functions which returns ordination and loadings. It could be simpler to add loadings to rowData() (easier to find them from there), however, I do not know how feasible the solutions is if there are many components (it might be ok). However, if the rest of results are stored in metadata/reducedDim, it might be confusing

antagomir commented 1 month ago

I think this can be merged. Loadings could also be in rowData but I think this we discussed already.

This same thing applies to other functions which returns ordination and loadings. It could be simpler to add loadings to rowData() (easier to find them from there), however, I do not know how feasible the solutions is if there are many components (it might be ok). However, if the rest of results are stored in metadata/reducedDim, it might be confusing

Most improtant that we are consistent. One problem with rowData is that the rowData columns should then be recognized by their name (or does the structure allow having multiple separate rowData elements.. or can rowData include lists..)

TuomasBorman commented 4 weeks ago

I think this can be merged. Loadings could also be in rowData but I think this we discussed already.

This same thing applies to other functions which returns ordination and loadings. It could be simpler to add loadings to rowData() (easier to find them from there), however, I do not know how feasible the solutions is if there are many components (it might be ok). However, if the rest of results are stored in metadata/reducedDim, it might be confusing

Most improtant that we are consistent. One problem with rowData is that the rowData columns should then be recognized by their name (or does the structure allow having multiple separate rowData elements.. or can rowData include lists..)

rowData can have any kind of object (I think) that has as many elements or rows as TreeSE has rows.

data("GlobalPatterns")
tse <- GlobalPatterns

# Does work
rowData(tse)[["asd"]] <- assay(tse)
rowData(tse)[["asd2"]] <- as.list(rownames(tse))

# Does not work
rowData(tse)[["asd"]] <- assay(tse) |> t()
rowData(tse)[["asd2"]] <- as.list(colnames(tse))