runEdgeR: Test for differential abundance: method 'treeAGG-DA-edgeR'
In markrobinsonuzh/treeAGG: Tree Aggregation

Description Usage Arguments Details Value Examples

Test differential abundance of entities using functions from the edgeR (Robinson et al. 2010, Bioinformatics; McCarthy et al. 2012, Nucleic Acids Research) to fit models and calculate moderated test for each entity. We have used estimateDisp to estimate the dispersion. The statistical methods implemented in the edgeR package were originally designed for the analysis of gene expression data such as RNA-sequencing counts. Here, we apply these methods to counts that might be from microbes or cells.

1
2
3

runEdgeR(obj, design = NULL, contrast = NULL, normalize = TRUE,
  method = "TMM", adjust.method = "BH", prior.count = 0.125,
  use.assays = NULL)

`obj`	A treeSummarizedExperiment object.
`design`	A numeric matrix. It must be of full column rank. Defaults to use all columns of `colData` to create design matrix. Note: Users should check whether the default created design matrix is exactly what they want or create their own design matrix using `model.matrix`.
`contrast`	numeric vector specifying one contrast of the linear model coefficients to be tested equal to zero. Its length must equal to the number of columns of design. If NULL, the last coefficient will be tested equal to zero.
`normalize`	A logical value, TRUE or FALSE. The default is TRUE.
`method`	Normalization method to be used. See `calcNormFactors` for more details.
`adjust.method`	A character string stating the method used to adjust p-values for multiple testing, passed on to `p.adjust`. It could be "bonferroni", "holm", "hochberg", "hommel", "BH", or "BY".
`prior.count`	average prior count to be added to observation to shrink the estimated log-fold-changes towards zero. See `prior.count` in `glmFit`
`use.assays`	A numeric vector. It specifies which matrix-like elements in assays will be used to do analysis.

The experimental design must be specified using a design matrix. The customized design matrix could be given by design.

Normalization for samples is automatically performed by edgeR package. More details about the calculation of normalization factor could be found from calcNormFactors. A sample might include entities corresponding to leaf nodes and internal nodes of tree. Only entities corresponding to leaf nodes are used to calculate the library size of each sample. The reason is that the abundance of an entity, corresponding to an internal node, is calculated by taking sum of the abundance from its descendant leaf nodes.

A treeSummarizedExperiment

`assays`	A list of tables
`rowData`	It stores the information of rows in `assays`, and the tables extracted from a `DGELRT` object that is generated by `glmLRT`. The later is stored as the internal part of the `rowData`. More details or example could be found in the vignette `Example of data analysis`
`colData`	NULL
`metadata`	`use.assays` which elements in the `assays` have been used to run differential abundance analysis. `design` the design matrix as input. `contrast` the contrast vector as input. `output_glmFit` the output from `glmFit`. A object of `DGEGLM-class`

library(S4Vectors)
set.seed(1)
y <- matrix(rnbinom(300,size=1,mu=10),nrow=10)
colnames(y) <- paste(rep(LETTERS[1:3], each = 10), rep(1:10,3), sep = "_")
rownames(y) <- tinyTree$tip.label

rowInf <- DataFrame(nodeLab = rownames(y),
                    var1 = sample(letters[1:3], 10, replace = TRUE),
                    var2 = sample(c(TRUE, FALSE), 10, replace = TRUE))
colInf <- DataFrame(gg = factor(sample(1:3, 30, replace = TRUE)),
                    group = rep(LETTERS[1:3], each = 10))
toy_lse <- leafSummarizedExperiment(tree = tinyTree, rowData = rowInf,
                                    colData = colInf,
                                    assays = list(y, (2*y), 3*y))

toy_tse <- nodeValue(data = toy_lse, fun = sum, tree = tinyTree,
message = TRUE)

# build the model
contrastList <- list(contrast1 = c(0, 0, 0, -1, 1),
                     contrast2 = c(0, -1, 1, 0, 0))
mod <- runEdgeR(obj = toy_tse, contrast = contrastList)
# show results gained from the second element of the assasy
# sort by PValue
topNodes(mod, sort.by = "PValue", use.assays = 2)