This is a wrapper function that delegates GSEA analyses to different "workers", each of which implements the flavor of GSEA of your choosing. The particular analyses that are performed are specified by the methods argument, and these methods are fine tuned by passing their arguments down through the ... of this wrapper function.

seas(
  x,
  gsd,
  methods = NULL,
  design = NULL,
  contrast = NULL,
  use.treat = FALSE,
  feature.min.logFC = if (use.treat) log2(1.25) else 1,
  feature.max.padj = 0.1,
  trim = 0.1,
  verbose = FALSE,
  ...,
  score.by = c("t", "logFC", "pval"),
  rank_by = NULL,
  rank_order = c("ordered", "descending", "ascending"),
  xmeta. = NULL,
  BPPARAM = BiocParallel::SerialParam()
)

Arguments

x

An object to run enrichment analyses over. This can be an ExpressoinSet-like object that you can differential expression over (for roast, fry, camera), a named (by feature_id) vector of scores to run ranked-based GSEA, a data.frame with feature_id's, ranks, scores, etc.

gsd

The GeneSetDb() that defines the gene sets of interest.

methods

A character vector indicating the GSEA methods you want to run. Refer to the GSEA Methods section for more details. If no methods are specified, only differential gene expression and geneset level statistics for the contrast are computed.

design

A design matrix for the study

contrast

The contrast of interest to analyze. This can be a column name of design, or a contrast vector which performs "coefficient arithmetic" over the columns of design. The design and contrast parameters are interpreted in exactly the same way as the same parameters in limma's limma::camera() and limma::roast() methods.

use.treat

should we use limma/edgeR's "treat" functionality for the gene-level differential expression analysis?

feature.min.logFC

The minimum logFC required for an individual feature (not geneset) to be considered differentialy expressed. Used in conjunction with feature.max.padj primarily for summarization of genesets (by geneSetsStats(), but can also be used by GSEA methods that require differential expression calls at the individual feature level, like goseq().

feature.max.padj

The maximum adjusted pvalue used to consider an individual feature (not geneset) to be differentially expressed. Used in conjunction with feature.min.logFC.

trim

The amount to trim when calculated trimmed t and logFC statistics for each geneset. This is passed down to the geneSetsStats() function.

verbose

make some noise during execution?

...

The arguments are passed down into calculateIndividualLogFC() and the various geneset analysis functions.

score.by

This tells us how to rank the features after differential expression analysis when x is an expression container. It specifies the name of the column to use downstream of a differential expression analysis over x. If x is a data.frame that needs to be ranked, see rank_by.

rank_by

Only works when x is a data.frame-like input. The name of a column that should be used to rank the features in x for pre-ranked gsea tests like cameraPR or fgsea. rank_by overrides score.by

rank_order

Only used when x is a data.frame-like input. Specifies how the features in x should be used to rank the features in x using the rank_by column. Accepted values are: "ordered" (default) means that the rows in x are pre-ranked already. "descendeing", and "ascending".

xmeta.

A hack to support data.frame inputs for x. End users should not use this.

BPPARAM

a BiocParallel parameter definition, like one generated from BiocParallel::MulticoreParam(), or BiocParallel::BatchtoolsParam(), for instance, which is passed down to BiocParallel::bplapply()]. Default is set to BiocParallel::SerialParam()

Value

A SparrowResult() which holds the results of all the analyses specified in the methods parameter.

Details

Set enrichment analyses can either be performed over an expression object, which requires the specification of the experiment design and contrast of interest, or over a set of features to rank (stored as a data.frame or vector).

Note that we are currently in the middle of a refactor to accept and fully take advantage of data.frame as inputs for x, which will be used for preranked type of GSEA methods. See the following issue for more details: https://github.com/lianos/multiGSEA/issues/24

The bulk of the GSEA methods currently available in this package come from edgeR/limma, however others are included (and are being added), as well. GSEA Methods and GSEA Method Parameterization sections for more details.

In addition to performing GSEA, this function also internally orchestrates a differential expression analysis, which can be tweaked by identifying the parameters in the calculateIndividualLogFC() function, and passing them down through ... here. The results of the differential expression analysis (ie. the limma::topTable()) are accessible by calling the logFC() function on the SparrowResult() object returned from this function call.

Please Note: be sure to cite the original GSEA method when using results generated from this function.

GSEA Methods

You can choose the methods you would like to run by providing a character vector of GSEA method names to the methods parameter. Valid methods you can select from include:

Methods annotated with a (*) indicate that these methods require a complete expression object, a valid design matrix, and a contrast specification in order to run. These are all of the same things you need to provide when performing a vanilla differential gene expression analysis.

Methods missing a (*) can be run on a feature-named input vector of gene level statistics which will be used for ranking (ie. a named vector of logFC's or t-statistics for genes). They can also be run by providing an expression, design, and contrast vector, and the appropriate statistics vector will be generated internally from the t-statistics, p-values, or log-fold-changes, depending on the value provided in the score.by parameter.

The worker functions that execute these GSEA methods are functions named do.METHOD within this package. These functions are not meant to be executed directly by the user, and are therefore not exported. Look at the respective method's help page (ie. if you are running "camera", look at the limma::camera() help page for full details. The formal parameters that these methods take can be passed to them via the ... in this seas() function.

GSEA Method Parameterization

Each GSEA method can be tweaked via a custom set of parameters. We leave the documentation of these parameters and how they affect their respective GSEA methods to the documentation available in the packages where they are defined. The seas() call simply has to pass these parameters down into the ... parameters here. The seas function will then pass these along to their worker functions.

What happens when two different GSEA methods have parameters with the same name?

Unfortunately you currently cannot provide different values for these parameters. An upcoming version version of sparrow will support this feature via slightly different calling semantics. This will also allow the caller to call the same GSEA method with different parameterizations so that even these can be compared against each other.

Differential Gene Expression

When the seas() call is given an expression matrix, design, and contrast, it will also internally orchestrate a gene level differential expression analysis. Depending on the type of expression object passed in via x, this function will guess on the best method to use for this analysis.

If x is a DGEList, then ensure that you have already called edgeR::estimateDisp() on x and edgeR's quasilikelihood framework will be used, otherwise we'll use limma (note that x can be an EList run through voom(), voomWithQuailityWeights(), or when where you have leveraged limma's limma::duplicateCorrelation() functionality, even.

The parameters of this differential expression analysis can also be customized. Please refer to the calculateIndividualLogFC() function for more information. The use.treat, feature.min.logFC, feature.max.padj, as well as the ... parameters from this function are passed down to that funciton.

Examples

vm <- exampleExpressionSet()
gdb <- exampleGeneSetDb()
mg <- seas(vm, gdb, c('camera', 'fry'),
           design = vm$design, contrast = 'tumor',
           # customzie camera parameter:
           inter.gene.cor = 0.04)
resultNames(mg)
#> [1] "camera" "fry"   
res.camera <- result(mg, 'camera')
res.fry <- result(mg, 'fry')
res.all <- results(mg)