eSetRevised 4 September, 2006 –
featureData slot. Revised 20 April 2007 – minor wording
changes; verbose and other arguments passed through
updateObject example; introduce a second variant of
initialize illustrating its use as a copy constructor.
Revised 4 November, 2010 – experimentData slot as
MIAxE class. MIAME class is now subclass of
MIAxE.
These notes help developers who are interested in using and
extending the eSet class hierarchy, and using features in
Biobase. The information here is not useful to regular users of
Biobase.
This document illustrates the Biobase structures and
approaches that make it it easy for developers to creatively use and
extend the eSet class hierarchy.
The document starts with a brief description of the motivation for
change, and a comparison of the old (before August, 2006) and new
eSets and related functionality (e.g.,the
Versioned class and updateObject methods). We then
illustrate how eSet can be extended to handle additional
types of data, and how new methods can exploit the eSet
class hierarchy. We conclude with a brief summary of lessons learned,
useful developer-related side-effects of efforts to revise
eSet, and possible directions for future development.
What is an eSet?
Motivation for change (August, 2006).
eSet to new
data types.Key features in the redesign.
updateObject methods (in the development branch).eSet object: high-throughput experimentsPurpose.
Structure: virtual base class.
## Virtual Class "eSet" [package "Biobase"]
##
## Slots:
##
## Name: assayData phenoData featureData
## Class: AssayData AnnotatedDataFrame AnnotatedDataFrame
##
## Name: experimentData annotation protocolData
## Class: MIAxE character AnnotatedDataFrame
##
## Name: .__classVersion__
## Class: Versions
##
## Extends:
## Class "VersionedBiobase", directly
## Class "Versioned", by class "VersionedBiobase", distance 2
##
## Known Subclasses: "ExpressionSet", "NChannelSet", "MultiSet", "SnpSet"assayData: high-throughput data.phenoData: sample covariates.featureData: feature covariates.experimentData: experimental description.annotation: assay description.?"eSet-class"assayData: high-throughput dataPurpose.
Structure: list, environment, or lockEnvironment class union.
?"AssayData-class"phenoData: sample covariatesPurpose.
Structure: AnnotatedDataFrame.
data: data.frame.
varMetadata: data.frame.
?"AnnotatedDataFrame-class"featureData: feature covariatesPurpose.
annotation slot for chip-level descriptions.Structure: AnnotatedDataFrame.
data: data.frame.
assayData.varMetadata: data.frame.
?"AnnotatedDataFrame-class"experimentData: experiment descriptionPurpose.
Structure: MIAxE virtual class
In Biobase version 2.11 the MIAxE virtual class was
introduced as a general container for experiment meta-data. The
MIAME class is now a subclass of MIAxE. See
?"MIAxE-class". The MIAME class structure is as
follows
title: experiment title.name: experimenter name(s).preprocessing: list of pre-processing steps.?"MIAME-class".annotation: assay descriptionPurpose.
Structure: character
eSet methodsInitialization.
eSet is VIRTUAL, initialize via subclass
callNextMethodAccessors (get, set).
assayData(obj); assayData(obj) <- value: access or
assign assayDataphenoData(obj); phenoData(obj) <- value: access or
assign phenoDataexperimentData(obj); experimentData(obj) <- value:
access or assign experimentDataannotation(obj); annotation(obj) <- value: access or
assign annotationSubsetting.
obj[i, j]: select genes i and samples j.obj$name; obj$name <- value: retrieve or assign
covariate name in phenoDataeSet methodsshow.storageMode: influence how assayData is
stored.updateObject: update eSet objects to their
current version.validObject: ensure that eSet is
valid.The validObject method is particularly important to
eSet, ensuring that eSet contains consistent
structure to data.
## function (object)
## {
## msg <- validMsg(NULL, isValidVersion(object, "eSet"))
## dims <- dims(object)
## if (ncol(dims) > 0) {
## msg <- validMsg(msg, assayDataValidMembers(assayData(object)))
## if (any(dims[1, ] != dims[1, 1]))
## msg <- validMsg(msg, "row numbers differ for assayData members")
## if (any(dims[2, ] != dims[2, 1]))
## msg <- validMsg(msg, "sample numbers differ for assayData members")
## if (dims[1, 1] != dim(featureData(object))[[1]])
## msg <- validMsg(msg, "feature numbers differ between assayData and featureData")
## if (!identical(featureNames(assayData(object)), featureNames(featureData(object))))
## msg <- validMsg(msg, "featureNames differ between assayData and featureData")
## if (dims[2, 1] != dim(phenoData(object))[[1]])
## msg <- validMsg(msg, "sample numbers differ between assayData and phenoData")
## if (!identical(sampleNames(assayData(object)), sampleNames(phenoData(object))))
## msg <- validMsg(msg, "sampleNames differ between assayData and phenoData")
## if (dim(phenoData(object))[[1]] != dim(protocolData(object))[[1]])
## msg <- validMsg(msg, "sample numbers differ between phenoData and protocolData")
## if (!identical(sampleNames(phenoData(object)), sampleNames(protocolData(object))))
## msg <- validMsg(msg, "sampleNames differ between phenoData and protocolData")
## }
## if (is.null(msg))
## TRUE
## else msg
## }
## <bytecode: 0x563325e90fc8>
## <environment: namespace:Biobase>The validity methods for eSet reflect our design goals.
All assayData members must have identical row and column
dimensions and featureNames. The names and numbers of
samples must be the same in assayData and
phenoData slots. Validity methods are defined for the
classes underlying each slot as well. For instance, the validity methods
for AnnotatedDataFrame check that variables used in
pData are at least minimally described in
varMetadata.
eSetBiobase defines three classes that extend eSet.
ExpressionSet (discussed further below) is meant to contain
microarray gene expression data. SnpSet is a preliminary class
to contain SNP data; other classes in development (e.g., in
oligo) may provide alternative implementations for SNP data.
MultiSet is an ExpressionSet-like class, but
without restriction on the names (though not structure) of elements in
the assayData slot.
ExpressionSetPurpose:
Required assayData members.
exprs, a matrix of expression values.Important methods.
Initialization (additional details below):
exprs(obj), exprs(obj) <- value: get or set
exprs; methods defined for ExpressionSet,
AssayData.
MultiSet.
assayData members: none.SnpSet.
assayData members: call,
callProbability.eSetA designer wanting to implement eSet for a particular
type of data creates a class that ‘contains’ eSet. The
steps for doing this are described below. One example of such a class is
ExpressionSet, designed to hold a matrix of gene expression
values in the assayData slot.
## Class "ExpressionSet" [package "Biobase"]
##
## Slots:
##
## Name: experimentData assayData phenoData
## Class: MIAME AssayData AnnotatedDataFrame
##
## Name: featureData annotation protocolData
## Class: AnnotatedDataFrame character AnnotatedDataFrame
##
## Name: .__classVersion__
## Class: Versions
##
## Extends:
## Class "eSet", directly
## Class "VersionedBiobase", by class "eSet", distance 2
## Class "Versioned", by class "eSet", distance 3## function (object)
## {
## msg <- validMsg(NULL, isValidVersion(object, "ExpressionSet"))
## msg <- validMsg(msg, assayDataValidMembers(assayData(object),
## c("exprs")))
## if (class(experimentData(object)) != "MIAME")
## msg <- validMsg(msg, "experimentData slot in ExpressionSet must be 'MIAME' object")
## if (is.null(msg))
## TRUE
## else msg
## }
## <bytecode: 0x563326325928>
## <environment: namespace:Biobase>The data structure of an ExpressionSet is identical to
that of eSet, and in fact is inherited (without additional
slot creation) from eSet. The main difference is that the
validity methods of eSet are augmented by a method to check
that the assayData slot contains an entity named
"exprs". A valid ExpressionSet object must
also satisfy all the validity requirements of eSet, but the
developer does not explicitly invoke validity checking of the parts of
the data structure inherited from eSet.
We want the Swirl data set (see the SW two
color data set that motivates this class) to contain four elements in
the assayData slot: R, G, Rb, Gb. To derive a
class from eSet for this data, we create a class, and
provide initialization and validation methods.
We create a class as follows:
Notice that there are no new data elements in SwirlSet
compared with eSet. The initialize method is
written as
setMethod("initialize", "SwirlSet",
function(.Object,
R = new("matrix"),
G = new("matrix"),
Rb = new("matrix"),
Gb = new("matrix"),
...) {
callNextMethod(.Object,
R=R, G=G, Rb=Rb, Gb=Gb,
...)
}) A slightly different initialize method allows the user
to specify either the assayData or the
assayData content. In advanced use, this has the advantage
that initialize can be used as a ‘copy constructor’ to
update several slots simultaneously.
setMethod("initialize", "SwirlSet",
function(.Object,
assayData=assayDataNew(
R=R, G=G, Rb=Rb, Gb=Gb),
R = new("matrix"),
G = new("matrix"),
Rb = new("matrix"),
Gb = new("matrix"),
...) {
if (!missing(assayData) &&
any(!missing(R), !missing(G), !missing(Rb), !missing(Gb))) {
warning("using 'assayData'; ignoring 'R', 'G', 'Rb', 'Gb'")
}
callNextMethod(.Object, assayData=assayData, ...)
})The structure of the initialize method is a bit
different from those often seen in R.
Often, initialize has only .Object as a named
argument, or, if there are other named arguments, they correspond to
slot names. Here our initialize method accepts four arguments, named
after the assayData elements. Inside the
initialize method, the named arguments are passed to the
next initialization method in the hierarchy (i.e.,
initialize defined for eSet). The
eSet initialize method then uses these
arguments to populate the data slots in .Object. In
particular, eSet places all arguments other
phenoData, experimentData, and
annotation into the assayData slot. The
eSet initialize method then returns the result
to the initialize method of SwirlSet, which
returns a SwirlSet object to the user:
General programming guidelines emerge from experience with the
initialize method of eSet and derived classes.
First, an appropriate strategy is to name only those data elements that
will be manipulated directly by the initialize method. For
instance, the definition above did not name
phenoData and other eSet slots by name. To do
so is not incorrect, but would require that they be explicitly named
(e.g., phenoData=phenoData) in the
callNextMethod code. Second, the arguments
R, G, Rb, Rg are present in the initialize
method to provide defaults consistent with object construction; the
‘full’ form of callNextMethod, replicating the named
arguments, is required in the version of R in which this class was developed. Third,
named arguments can be manipulated before callNextMethod is
invoked. Fourth, the return value of callNextMethod can be
captured…
setMethod("initialize", "MySet",
function(.Object, ...) {
.Object <- callNextMethod(.Object, ...)
})and manipulated before being returned to the user. Finally, it is the
responsibility of the developer to ensure that a valid object is
created; callNextMethod is a useful way to exploit
correctly designed initialize methods for classes that the
object extends, but the developer is free to use other techniques to
create valid versions of their class.
A validity method might complete our new class. A validity method is
essential to ensure that the unique features of SwirlSet – our
reason for designing the new class – are indeed present. We define our
validity method to ensure that the assayData slot contains
our four types of expression elements:
setValidity("SwirlSet", function(object) {
assayDataValidMembers(assayData(object), c("R", "G", "Rb", "Gb"))
})## Class "SwirlSet" [in ".GlobalEnv"]
##
## Slots:
##
## Name: assayData phenoData featureData
## Class: AssayData AnnotatedDataFrame AnnotatedDataFrame
##
## Name: experimentData annotation protocolData
## Class: MIAxE character AnnotatedDataFrame
##
## Name: .__classVersion__
## Class: Versions
##
## Extends:
## Class "eSet", directly
## Class "VersionedBiobase", by class "eSet", distance 2
## Class "Versioned", by class "eSet", distance 3Notice that we do not have to explicitly request that the validity of other parts of the SwirlSet object are valid; this is done for us automatically. Objects are checked for validity when they are created, but not when modified. This is partly for efficiency reasons, and partly because object updates might transiently make them invalid. So a good programing practice is to ensure validity after modification, e.g.,
myFancyFunction <- function(obj) {
assayData(obj) <- fancyAssaydData # obj invalid...
phenoData(obj) <- justAsFancyPhenoData # but now valid
validObject(obj)
(obj)
}Assigning fancyAssaydData might invalidate the object,
but justAsFancyPhenoData restores validity.
One problem encountered in the Bioconductor project is that data
objects stored to disk become invalid as the underlying class definition
changes. For instance, earlier releases of Biobase contain a
sample eSet object. But under the changes discussed here,
eSet is virtual and the stored object is no longer valid.
The challenge is to easily identify invalid objects, and to provide a
mechanism for updating old objects to their new representation.
Biobase introduces the Versioned and
VersionedBiobase classes to facilitate this. These classes are
incorporated into key Biobaseclass definitions.Biobase
also defines updateObject methods (the
updateObject generic function is defined in the BiocGenerics
package) for conveniently updating old objects to their new
representation.
## R Biobase eSet ExpressionSet
## "2.13.0" "2.11.5" "1.3.0" "1.0.0"The version information for this object is a named list. The first two elements indicate the version of R and Biobase used to create the object. The latter two elements are contained in the class prototype, and the class prototype is consulted to see if the instance of an object is ‘current’. These lists can be subsetted in the usual way, e.g.,
## eSet ExpressionSet
## TRUE TRUEVersioned classes, updateObject and related methods
simplify the long-term maintenance of data objects. Take the fictious
MySet as an example.
setClass("MySet",
contains = "eSet",
prototype = prototype(new("VersionedBiobase",
versions=c(classVersion("eSet"), MySet="1.0.0"))))
obj <- new("MySet")
classVersion(obj)## R Biobase eSet MySet
## "4.4.1" "2.67.0" "1.3.0" "1.0.0"This is a new class, and might undergo changes in its structure at some point in the future. When these changes are introduced, the developer will change the version number of the class in its prototype (the last line, below):
setClass("MySet",
contains = "eSet",
prototype = prototype(
new("VersionedBiobase",
versions=c(classVersion("eSet"), MySet="1.0.1"))))
isCurrent(obj)## S4 R Biobase eSet MySet
## TRUE TRUE TRUE TRUE FALSEand add code to update to the new version
setMethod("updateObject", signature(object="MySet"),
function(object, ..., verbose=FALSE) {
if (verbose) message("updateObject(object = 'MySet')")
object <- callNextMethod()
if (isCurrent(object) ["MySet"]) return(object)
## Create an updated instance.
if (!isVersioned(object))
## Radical surgery – create a new, up-to-date instance
new("MySet",
assayData = updateObject(assayData(object),
..., verbose=verbose),
phenoData = updateObject(phenoData(object),
..., verbose=verbose),
experimentData = updateObject(experimentData(object),
..., verbose=verbose),
annotation = updateObject(annotation(object),
..., verbose=verbose))
else {
## Make minor changes, and update version by consulting class definition
classVersion(object)["MySet"]<-
classVersion("MySet")["MySet"]
object
}
})The code after if(!isVersioned) illustrates one way of
performing ’radical surgery, creating a new up-to-date instance by
updating all slots. The else clause represents more modest
changes, using methods to update stale information.
updateObject then returns a new, enhanced object:
## R Biobase eSet MySet
## "4.4.1" "2.67.0" "1.3.0" "1.0.1"As in the example, versioning helps in choosing which modifications to perform – minor changes for a slightly out-of-date object, radical surgery for something more ancient. Version information might also be used in methods, where changing class representation might facilitate more efficient routines.
The information on R and
Biobase versions is present in eSet derived
classes because eSet contains VersionedBiobase. On
the other hand, AnnotatedDataFrame contains Versioned,
and has only information about its own class version.
## AnnotatedDataFrame
## "1.1.0"The rationale for this is that AnnotatedDataFrame is and
will likely remain relatively simple, and details about R and Biobase are probably irrelevant
to its use. On the other hand, some aspects of eSet and the
algorithms that operate on them are more cutting edge and subject to
changes in R or Biobase.
Knowing the version of R and
Biobase used to create an instance might provide valuable
debugging information.
The key to versioning your own classes is to define your class to
contain Versioned or VersionedBiobase,
and to add the version information in the prototype. For instance, to
add a class-specific version stamp to SwirlSet we would modify
the class definition to
setClass("SwirlSet", contains = "eSet",
prototype = prototype(
new("VersionedBiobase",
versions=c(classVersion("eSet"), SwirlSet="1.0.0"))))
classVersion(new("SwirlSet"))## R Biobase eSet SwirlSet
## "4.4.1" "2.67.0" "1.3.0" "1.0.0"See additional examples in the Versioned help page.
It is also possible to add arbitrary information to particular instances.
## R Biobase eSet SwirlSet MyID
## "4.4.1" "2.67.0" "1.3.0" "1.0.0" "0.0.1"## R Biobase eSet SwirlSet MyID
## "4.4.1" "2.67.0" "1.3.0" "1.0.0" "0.0.1"There is additional documentation about these classes and methods in Biobase.
This document summarizes Biobase, outlining strategies that
developers using Biobase may find useful. The main points are
to introduce the eSet class hierarchy, to illustrate how
developers can effectively extend this class, and to introduce class
versions as a way of tracking and easily updating objects. It is
anticipated that eSet-derived classes will play an
increasingly important role in Biobase development.
The version number of R and packages loaded for generating the vignette were:
## R version 4.4.1 (2024-06-14)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.1 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so; LAPACK version 3.12.0
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## time zone: Etc/UTC
## tzcode source: system (glibc)
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] Biobase_2.67.0 BiocGenerics_0.53.1 generics_0.1.3
## [4] BiocStyle_2.35.0
##
## loaded via a namespace (and not attached):
## [1] digest_0.6.37 R6_2.5.1 fastmap_1.2.0
## [4] xfun_0.48 maketools_1.3.1 cachem_1.1.0
## [7] knitr_1.48 htmltools_0.5.8.1 rmarkdown_2.28
## [10] buildtools_1.0.0 lifecycle_1.0.4 cli_3.6.3
## [13] sass_0.4.9 jquerylib_0.1.4 compiler_4.4.1
## [16] sys_3.4.3 tools_4.4.1 evaluate_1.0.1
## [19] bslib_0.8.0 yaml_2.3.10 BiocManager_1.30.25
## [22] jsonlite_1.8.9 rlang_1.1.4
Comments on
assayData: high-throughput data storageThe
assayDataslot is meant to store high-throughput data. The idea is that the slot contains identically sized matrices containing expression or other data. All matrices in the slot must have the same dimension, and are structured so that rows represent ‘features’ and columns represent ‘samples’. Validity methods enforce that row and column names of slot elements are identical.For technical reasons, creating instance of AssayData is slightly different from the way this is usually done in R. Normally, one creates an instance of a class with an expression like
new("ExpressionSet", ...), with the … representing additional arguments. AssayData objects are created withwhere
eltmight be a matrix of expression values. For the curious, the reason for this setup stems from our desire to have a class thatisa list or environment, rather than a class that has a slot that contains a list or environment. Theisrelationship is desirable to avoid unnecessary function calls to access slots, and requires that a classcontainthe base type (e.g., environment). Until recently an R object could notcontainan environment.The
assayDataslot ofExpressionSetobjects must contain a matrix namedexprs. Nonetheless, theExpressionSetvalidity method tries to be liberal – it guarantees that the object has anexprselement, but allows for other elements too. The prudent developer wanting consistent additional data elements should derive a class fromExpressionSetthat enforces the presence of their desired elements.The AssayData class allows for data elements to be stored in three different ways (see
?storageModeand?"storageMode<-"for details): as alist,environment, orlockedEnvironment. Developers are probably familiar withlistobjects; a drawback is thatexprselements may be large, and some operations on lists in R may trigger creation of many copies of the theexprselement. This can be expensive in both space and time. Environments are nearly unique in R, in that they are passed by reference rather than value. This eliminates some copying, but has the unfortunate consequence that side-effects occur – modifications to an environment inside a function influence the value of elements outside the function. For these reasons, environments can be useful as ‘read only’ arguments to functions, but can have unexpected consequences when functions modify their arguments. Locked environments implemented in Biobase try to strike a happy medium, allowing pass by reference for most operations but triggering (whole-environment) copying when elements in the environment are modified. The locking mechanism is enforced by only allowing known ‘safe’ operations to occur, usually by channeling user actions through the accessor methods:The
setReplaceMethodforexprs(andassayData) succeeds by performing a deep copy of the entire environment. Because this is very inefficient, the recommended paradigm to update an element in alockedEnvironmentis to extract it, make many changes, and then reassign it. Developers can studyassayDatamethods to learn more about how to lock and unlock environment bindings. Biobase allows the experienced user to employ (and run the risks of) environments, but the expectation is that most user objects are constructed with the defaultlockedEnvironmentorlist.A longer term consideration in designing AssayData was to allow more flexible methods of data storage, e.g., through database-hosted arrays. This is facilitated by using generic functions such as
exprs()for data access, so that classes derived from AssayData can provide implementations appropriate for their underlying storage mode.