SummarizedExperiment-class: SummarizedExperiment objects

SummarizedExperiment-classR Documentation

SummarizedExperiment objects

Description

The SummarizedExperiment class is a matrix-like container where rows represent features of interest (e.g. genes, transcripts, exons, etc...) and columns represent samples (with sample data summarized as a DataFrame). A SummarizedExperiment object contains one or more assays, each represented by a matrix-like object of numeric or other mode.

Note that SummarizedExperiment is the parent of the RangedSummarizedExperiment class which means that all the methods documented below also work on a RangedSummarizedExperiment object.

Usage


## Constructor

# See ?RangedSummarizedExperiment for the constructor function.

## Accessors

assayNames(x, ...)
assayNames(x, ...) <- value
assays(x, withDimnames=TRUE, ...)
assays(x, withDimnames=TRUE, ...) <- value
assay(x, i, withDimnames=TRUE, ...)
assay(x, i, withDimnames=TRUE, ...) <- value
rowData(x, use.names=TRUE, ...)
rowData(x, ...) <- value
colData(x, ...)
colData(x, ...) <- value
#dim(x)
#dimnames(x)
#dimnames(x) <- value

## Quick colData access

## S4 method for signature 'SummarizedExperiment'
x$name
## S4 replacement method for signature 'SummarizedExperiment'
x$name <- value
## S4 method for signature 'SummarizedExperiment,ANY,missing'
x[[i, j, ...]]
## S4 replacement method for signature 'SummarizedExperiment,ANY,missing'
x[[i, j, ...]] <- value

## Subsetting

## S4 method for signature 'SummarizedExperiment'
x[i, j, ..., drop=TRUE]
## S4 replacement method for signature 'SummarizedExperiment,ANY,ANY,SummarizedExperiment'
x[i, j] <- value
## S4 method for signature 'SummarizedExperiment'
subset(x, subset, select, ...)

## Combining

## S4 method for signature 'SummarizedExperiment'
rbind(..., deparse.level=1)
## S4 method for signature 'SummarizedExperiment'
cbind(..., deparse.level=1)
## S4 method for signature 'SummarizedExperiment'
combineRows(x, ..., delayed=TRUE, fill=NA, use.names=TRUE)
## S4 method for signature 'SummarizedExperiment'
combineCols(x, ..., delayed=TRUE, fill=NA, use.names=TRUE)

## On-disk realization
## S4 method for signature 'SummarizedExperiment'
realize(x, BACKEND=getAutoRealizationBackend())

Arguments

x

A SummarizedExperiment object.

...

For assay, arguments in ... are forwarded to assays.

For rbind, cbind, ... contains SummarizedExperiment objects to be combined.

For other accessors, ignored.

value

An object of a class specified in the S4 method signature or as outlined in ‘Details’.

i, j

For assay, assay<-, i is an integer or numeric scalar; see ‘Details’ for additional constraints.

For [,SummarizedExperiment, [,SummarizedExperiment<-, i, j are subscripts that can act to subset the rows and columns of x, that is the matrix elements of assays.

For [[,SummarizedExperiment, [[<-,SummarizedExperiment, i is a scalar index (e.g., character(1) or integer(1)) into a column of colData.

name

A symbol representing the name of a column of colData.

withDimnames

A logical(1), indicating whether the dimnames of the SummarizedExperiment object should be applied (i.e. copied) to the extracted assays. More precisely, setting withDimnames=FALSE in the getter returns the assays as-is whereas setting withDimnames=FALSE return them with possibly modified dimnames.

Setting withDimnames=FALSE in the setter (assays<-) is required when the dimnames on the supplied assays are not identical to the dimnames on the SummarizedExperiment object; it does not influence actual assignment of dimnames to assays (they're always stored as-is).

Note that

  assays(x, withDimnames=FALSE) <- assays(x, withDimnames=FALSE)

is guaranteed to always work and be a no-op. This is not the case if withDimnames=TRUE is used or if withDimnames is not specified.

use.names

For rowData: Like mcols(x), by default rowData(x) propagates the rownames of x to the returned DataFrame object (note that for a SummarizedExperiment object, the rownames are also the names i.e. rownames(x) is always the same as names(x)). Setting use.names=FALSE suppresses this propagation i.e. it returns a DataFrame object with no rownames. Use this when rowData(x) fails, which can happen when the rownames contain NAs (because the rownames of a SummarizedExperiment object can contain NAs, but the rownames of a DataFrame object cannot).

For combineRows and combineCols: See Combining section below.

drop

A logical(1), ignored by these methods.

deparse.level

See ?base::cbind for a description of this argument.

subset

An expression which, when evaluated in the context of rowData(x), is a logical vector indicating elements or rows to keep: missing values are taken as false.

select

An expression which, when evaluated in the context of colData(x), is a logical vector indicating elements or rows to keep: missing values are taken as false.

delayed, fill

See combineRows and combineCols in Combining section below.

BACKEND

NULL (the default), or a single string specifying the name of the backend. When the backend is set to NULL, each element of assays(x) is realized in memory as an ordinary array by just calling as.array on it.

Details

The SummarizedExperiment class is meant for numeric and other data types derived from a sequencing experiment. The structure is rectangular like a matrix, but with additional annotations on the rows and columns, and with the possibility to manage several assays simultaneously so long as they be of the same dimensions.

The rows of a SummarizedExperiment object represent features of interest. Information about these features is stored in a DataFrame object, accessible using the function rowData. The DataFrame must have as many rows as there are rows in the SummarizedExperiment object, with each row of the DataFrame providing information on the feature in the corresponding row of the SummarizedExperiment object. Columns of the DataFrame represent different attributes of the features of interest, e.g., gene or transcript IDs, etc.

Each column of a SummarizedExperiment object represents a sample. Information about the samples are stored in a DataFrame, accessible using the function colData, described below. The DataFrame must have as many rows as there are columns in the SummarizedExperiment object, with each row of the DataFrame providing information on the sample in the corresponding column of the SummarizedExperiment object. Columns of the DataFrame represent different sample attributes, e.g., tissue of origin, etc. Columns of the DataFrame can themselves be annotated (via the mcols function). Column names typically provide a short identifier unique to each sample.

A SummarizedExperiment object can also contain information about the overall experiment, for instance the lab in which it was conducted, the publications with which it is associated, etc. This information is stored as a list object, accessible using the metadata function. The form of the data associated with the experiment is left to the discretion of the user.

The SummarizedExperiment container is appropriate for matrix-like data. The data are accessed using the assays function, described below. This returns a SimpleList object. Each element of the list must itself be a matrix (of any mode) and must have dimensions that are the same as the dimensions of the SummarizedExperiment in which they are stored. Row and column names of each matrix must either be NULL or match those of the SummarizedExperiment during construction. It is convenient for the elements of SimpleList of assays to be named.

Constructor

SummarizedExperiment instances are constructed using the SummarizedExperiment function documented in ?RangedSummarizedExperiment.

Accessors

In the following code snippets, x is a SummarizedExperiment object.

assays(x), assays(x) <- value:

Get or set the assays. value is a list or SimpleList, each element of which is a matrix with the same dimensions as x.

assay(x, i), assay(x, i) <- value:

A convenient alternative (to assays(x)[[i]], assays(x)[[i]] <- value) to get or set the ith (default first) assay element. value must be a matrix of the same dimension as x, and with dimension names NULL or consistent with those of x.

assayNames(x), assayNames(x) <- value:

Get or set the names of assay() elements.

rowData(x, use.names=TRUE), rowData(x) <- value:

Get or set the row data. value is a DataFrame object.

colData(x), colData(x) <- value:

Get or set the column data. value is a DataFrame object. Row names of value must be NULL or consistent with the existing column names of x.

metadata(x), metadata(x) <- value:

Get or set the experiment data. value is a list with arbitrary content.

dim(x):

Get the dimensions (features of interest x samples) of the SummarizedExperiment.

dimnames(x), dimnames(x) <- value:

Get or set the dimension names. value is usually a list of length 2, containing elements that are either NULL or vectors of appropriate length for the corresponding dimension. value can be NULL, which removes dimension names. This method implies that rownames, rownames<-, colnames, and colnames<- are all available.

Subsetting

In the code snippets below, x is a SummarizedExperiment object.

x[i,j], x[i,j] <- value:

Create or replace a subset of x. i, j can be numeric, logical, character, or missing. value must be a SummarizedExperiment object with dimensions, dimension names, and assay elements consistent with the subset x[i,j] being replaced.

subset(x, subset, select):

Create a subset of x using an expression subset referring to columns of rowData(x) and / or select referring to column names of colData(x).

Additional subsetting accessors provide convenient access to colData columns

x$name, x$name <- value

Access or replace column name in x.

x[[i, ...]], x[[i, ...]] <- value

Access or replace column i in x.

Combining

In the code snippets below, x, y and ... are SummarizedExperiment objects to be combined.

rbind(...):

rbind combines objects with the same samples but different features of interest (rows in assays). The colnames in rowData(SummarizedExperiment) must match or an error is thrown. Duplicate columns of colData(SummarizedExperiment) must contain the same data.

Data in assays are combined by name matching; if all assay names are NULL matching is by position. A mixture of names and NULL throws an error.

metadata from all objects are combined into a list with no name checking.

cbind(...):

cbind combines objects with the same features of interest but different samples (columns in assays). The colnames in colData(SummarizedExperiment) must match or an error is thrown. Duplicate columns of rowData(SummarizedExperiment) must contain the same data.

Data in assays are combined by name matching; if all assay names are NULL matching is by position. A mixture of names and NULL throws an error.

metadata from all objects are combined into a list with no name checking.

combineRows(x, ..., use.names=TRUE, delayed=TRUE, fill=NA):

combineRows acts like more flexible rbind, returning a SummarizedExperiment with features equal to the concatenation of features across all input objects. Unlike rbind, it permits differences in the number and identity of the columns, differences in the available rowData fields, and even differences in the available assays among the objects being combined.

If use.names=TRUE, each input object must have non-NULL, non-duplicated column names. These names do not have to be the same, or even shared, across the input objects. The column names of the returned SummarizedExperiment will be a union of the column names across all input objects. If a column is not present in an input, the corresponding assay and colData entries will be filled with fill and NAs, respectively, in the combined SummarizedExperiment.

If use.names=FALSE, all objects must have the same number of columns. The column names of the returned object is set to colnames(x). Any differences in the column names between input objects are ignored.

Data in assays are combined by matching the names of the assays. If one input object does not contain a named assay present in other input objects, the corresponding assay entries in the returned object will be set to fill. If all assay names are NULL, matching is done by position. A mixture of named and unnamed assays will throw an error.

If delayed=TRUE, assay matrices are wrapped in DelayedArrays to avoid any extra memory allocation during the matrix rbinding. Otherwise, the matrices are combined as-is; note that this may still return DelayedMatrixs if the inputs were also DelayedMatrix objects.

If any input is a RangedSummarizedExperiment, the returned object will also be a RangedSummarizedExperiment. The rowRanges of the returned object is set to the concatenation of the rowRanges of all inputs. If any input is a SummarizedExperiment, the returned rowRanges is converted into a GRangesList and the entries corresponding to the rows of the SummarizedExperiment are set to zero-length GRanges. If all inputs are SummarizedExperiment objects, a SummarizedExperiment is also returned.

rowData are combined using combineRows for DataFrame objects. It is not necessary for all input objects to have the same fields in their rowData; missing fields are filled with NAs for the corresponding rows in the returned object.

metadata from all objects are combined into a list with no name checking.

combineCols(x, ..., use.names=TRUE, delayed=TRUE, fill=NA):

combineCols acts like more flexible cbind, returning a SummarizedExperiment with columns equal to the concatenation of columns across all input objects. Unlike cbind, it permits differences in the number and identity of the rows, differences in the available colData fields, and even differences in the available assays among the objects being combined.

If use.names=TRUE, each input object must have non-NULL, non-duplicated row names. These names do not have to be the same, or even shared, across the input objects. The row names of the returned SummarizedExperiment will be a union of the row names across all input objects. If a row is not present in an input, the corresponding assay and rowData entries will be filled with fill and NAs, respectively, in the combined SummarizedExperiment.

If use.names=FALSE, all objects must have the same number of rows. The row names of the returned object is set to rownames(x). Any differences in the row names between input objects are ignored.

Data in assays are combined by matching the names of the assays. If one input object does not contain a named assay present in other input objects, the corresponding assay entries in the returned object will be set to fill. If all assay names are NULL, matching is done by position. A mixture of named and unnamed assays will throw an error.

If delayed=TRUE, assay matrices are wrapped in DelayedArrays to avoid any extra memory allocation during the matrix rbinding. Otherwise, the matrices are combined as-is; note that this may still return DelayedMatrixs if the inputs were also DelayedMatrix objects.

If any input is a RangedSummarizedExperiment, the returned object will also be a RangedSummarizedExperiment. The rowRanges of the returned object is set to a merge of the rowRanges of all inputs, where the coordinates for each row are taken from the input object that contains that row. Any conflicting ranges for shared rows will raise a warning and all rowRanges information from the offending RangedSummarizedExperiment will be ignored. If any input is a SummarizedExperiment, the returned rowRanges is converted into a GRangesList and the entries corresponding to the unique rows of the SummarizedExperiment are set to zero-length GRanges. If all inputs are SummarizedExperiment objects, a SummarizedExperiment is also returned.

colData are combined using combineRows for DataFrame objects. It is not necessary for all input objects to have the same fields in their colData; missing fields are filled with NAs for the corresponding columns in the returned object.

metadata from all objects are combined into a list with no name checking.

Implementation and Extension

This section contains advanced material meant for package developers.

SummarizedExperiment is implemented as an S4 class, and can be extended in the usual way, using contains="SummarizedExperiment" in the new class definition.

In addition, the representation of the assays slot of SummarizedExperiment is as a virtual class Assays. This allows derived classes (contains="Assays") to implement alternative requirements for the assays, e.g., backed by file-based storage like NetCDF or the ff package, while re-using the existing SummarizedExperiment class without modification. See Assays for more information.

Author(s)

Martin Morgan; combineRows and combineCols by Aaron Lun

See Also

  • RangedSummarizedExperiment objects.

  • DataFrame, SimpleList, and Annotated objects in the S4Vectors package.

  • The metadata and mcols accessors in the S4Vectors package.

  • saveHDF5SummarizedExperiment and loadHDF5SummarizedExperiment in the HDF5Array package for saving/loading an HDF5-based SummarizedExperiment object to/from disk.

  • The realize generic function in the DelayedArray package for more information about on-disk realization of objects carrying delayed operations.

Examples

nrows <- 200; ncols <- 6
counts <- matrix(runif(nrows * ncols, 1, 1e4), nrows)
colData <- DataFrame(Treatment=rep(c("ChIP", "Input"), 3),
                     row.names=LETTERS[1:6])
se0 <- SummarizedExperiment(assays=SimpleList(counts=counts),
                            colData=colData)
se0
dim(se0)
dimnames(se0)
assayNames(se0)
head(assay(se0))
assays(se0) <- endoapply(assays(se0), asinh)
head(assay(se0))

rowData(se0)
colData(se0)

se0[, se0$Treatment == "ChIP"]
subset(se0, select = Treatment == "ChIP")

## rbind() combines objects with the same samples but different
## features of interest:
se1 <- se0
se2 <- se1[1:50,]
rownames(se2) <- letters[seq_len(nrow(se2))]
cmb2 <- rbind(se1, se2)
dim(cmb2)
dimnames(cmb2)

## cbind() combines objects with the same features of interest
## but different samples:
se1 <- se0
se2 <- se1[,1:3]
colnames(se2) <- letters[seq_len(ncol(se2))]
cmb1 <- cbind(se1, se2)
dim(cmb1)
dimnames(cmb1)

## ---------------------------------------------------------------------
## ON-DISK REALIZATION
## ---------------------------------------------------------------------
library(DelayedArray)
setAutoRealizationBackend("HDF5Array")
cmb3 <- realize(cmb2)
assay(cmb3, withDimnames=FALSE)  # an HDF5Matrix object

Bioconductor/SummarizedExperiment documentation built on Nov. 4, 2022, 11:59 a.m.