Benchmark against established annotation approaches

[thomazbastiaanssen]

Background

Initial version of mapModules has been built.

Validity of output

• Compare mapModules output vs current golden standard, rpm().
• implement testthat
  • unstratified case.
  • stratified case.

Functionality

• Currently, computes one of: "coverage" "presence" "sum" "count".
• Compare to rpm().

Code Quality

• Potentially simplify internals of mapModules.
  • stratified vs non-stratified can be simplified. At some point, I use split.data.frame (confusingly) to split stratified input x into strata/(taxa). Then, I proceed using lapply by stratum -> do.call(rbind, res_list).
    • simplification: surely unstratified can be seen as a single instance of the stratified analysis.

Bioconductor integration

• implement (t)se methods
• Should mapModules dispatch on feature table x rather than module mapping information map?
  • Current x was taken directly from a SummarizedExperiment, namely: x <- assay(SE)
  • Potentially even double dispatch, on both x, map?
    • Seems needlessly complicated - I don't expect to see non-MultiFactor methods for map in mapModules.

[RiboRings]

Code moved to https://github.com/Minotau-R/ariadne/blob/v2/R/weave_complex.R

We should check that matrix operations return expected results.

[RiboRings]

Comparison to humann and omixer here: #42

Conclusions:

• humann output matches
• omixer output matches for coverage (except for omixer comma bug) but differs for abundance, because omixer abundance estimation is based on maximising abundance for non-redundant steps in the pathway: https://github.com/raeslab/omixer-rpm/tree/master#description

But ariadne main output in principle is coverage, not abundances, so that's ok. Custom abundance estimation is implemented in mia with agglomerateByModule and applyByModule.