 CITE: theseus R package https://www.biorxiv.org/content/biorxiv/early/2018/04/05/295675.full.pdf

- mare for preprocessing, significance tests (continuous and discrete;
  neg binomial; with some standard covariates) and analysis workflow;
  very useful tools but implementation is half-way

- PathoStat has a Shiny and number of features already implemented in R and
  operating with phyloseq that can be utilized: PCoA, UniFrac,
  Exploratory Trees, Limma, DeSeq2, time series visualizations. The
  core and diversity stuff is less developed than in microbiome
  package.

- structSSI for hierarchical testing should be used

- treeDA provides a more rigorous replacement for LEFSe (sparsity +
  discrimnant analysis to identify differential taxa)

- agPCA for phylogeny aware PCA

- phylofactor for phylogeny aware factor analysis

-> Many of these could also have Bayesian variants. Start with
   hierarchical testing procedures.

otu arcsin-square root transformed abundance transformation for OTU abundances -> used at least in Huttenhower MaAsLin method. What is the advantage?

- Alternatives to EMD could be provided by indicator species and phi-coefficient analysis that have has been used to identify OTUs associated with disease states `r citep(bib[["Griffin2017"]])`. The indicator species analysis quantifies the association between the given taxonomic group and condition based on a score derived from the occurrence probability and its specificity mean abundance in the condition `r citep(bib[["Dufrene1997"]])`. The Phi-coefficient analysis measures the correlation between the presence or absence of the taxonomic group and the given condition `r citep(bib[["Griffin2017"]])`. The community indicator value (CIV) quantifies how skewed the indicator species are towards a given condition `r citep(bib[["Griffin2017"]])`. The recent development of such metric highlights the need to develop measures of community differences that can identify responder subsets from a wider population `r citep(bib[["Salonen2014diethealth"]])`.

- Use variance stabilizing transformation instead of log (DEseq2); how
  relates to Hellinger? Add to microbiome transformations
  ps_dds <- phyloseq_to_deseq2(ps, ~ age_binned + family_relationship)
  varianceStabilizingTransformation(ps_dds, blind = TRUE, fitType = "parametric")
  ps_dds <- estimateSizeFactors(ps_dds)
  ps_dds <- estimateDispersions(ps_dds)
  abund <- getVarianceStabilizedData(ps_dds)

- structSSI: Multiple Testing for Hypotheses with Hierarchical or Group Structure



Gini index and wealth coefficient from Sudarshan-2016 ?

rdryad and phyloseq depend on RJSONIO:
library("tools")
dependsOnPkgs("RJSONIO")
This causes the BioC warnings in build/check
Could be replaced by custom copies of the necessary functions to get rid of this. Postponing for now.





if (probe.parameters == "frpa.HITChip") {
    if (verbose) {message("Loading pre-calculated preprocessing parameters")}
    rpa.hitchip.species.probe.parameters <- list()
    load(system.file("extdata/probe.parameters.rda", package = "HITChipDB"))
    probe.parameters <- rpa.hitchip.species.probe.parameters

    # Ensure we use only those parameters that are in the filtered phylogeny
    for (bac in names(probe.parameters)) {
      probe.parameters[[bac]] <- probe.parameters[[bac]][intersect(names(probe.parameters[[bac]]), probesets[[bac]])]
    }
  } else if (is.null(probe.parameters)) {
    probe.parameters <- list()
  }



