---
title: "Creating new `MsBackend` classes"
output:
    BiocStyle::html_document:
        toc_float: true
vignette: >
    %\VignetteIndexEntry{Creating new `MsBackend` class}
    %\VignetteEngine{knitr::rmarkdown}
    %\VignetteEncoding{UTF-8}
    %\VignettePackage{Spectra}
    %\VignetteDepends{Spectra,BiocStyle}
bibliography: references.bib
---

```{r style, echo = FALSE, results = 'asis', message=FALSE}
BiocStyle::markdown()
```

**Package**: `r Biocpkg("Spectra")`<br />
**Authors**: `r packageDescription("Spectra")[["Author"]] `<br />
**Last modified:** `r file.info("Spectra.Rmd")$mtime`<br />
**Compiled**: `r date()`

```{r, echo = FALSE, message = FALSE}
library(Spectra)
library(BiocStyle)
```

# Introduction

This vignette briefly describes the `MsBackend` class which is used by the
`Spectra` package to *represent* and provide Mass Spectrometry (MS) data and
illustrates how a new such *backend* class can be created and tested for
validity.

Contributions to this vignette (content or correction of typos) or requests for
additional details and information are highly welcome (ideally *via* pull
requests or github issues).


# What is a `MsBackend`?

The `Spectra` package separates the code for the analysis of MS data from the
code needed to import, represent and provide the data. The former is implemented
for the `Spectra` class which is the main object users will use for their
analyses. The `Spectra` object relies on a so-called *backend* to provide the MS
data. The `MsBackend` virtual class defines the API that new *backend* classes
need to implement in order to be used with the `Spectra` object. Each `Spectra`
object contains an implementation of such a `MsBackend` within its `@backend`
slot which provides the MS data to the `Spectra` object. All data management is
thus hidden from the user. In addition this separation allows to define new,
alternative, data representations and integrate them seamlessly into a
`Spectra`-based data analysis workflow.

This concept is an extension of the of *in-memory* and *on-disk* data
representations from the `r Biocpkg("MSnbase")` package
[@gattoMSnbaseEfficientElegant2020a].


## Conventions and definitions

General conventions for MS data of a `Spectra` are:

- One `Spectra` object is supposed to contain MS (spectral) data of multiple
  MS spectra.
- m/z values within each spectrum are expected to be sorted increasingly.
- Missing values (`NA`) for m/z values are not supported.
- Properties of a spectrum are called *spectra variables*. While backends can
  define their own properties, a minimum required set of spectra variables
  **must** be provided by each backend (even if their values are empty). These
  *core spectra variables* are listed (along with their expected data type) by
  the `coreSpectraVariables()` function.
- `dataStorage` and `dataOrigin` are two special spectra variables that define
  for each spectrum where the data is stored and from where the data derived (or
  was loaded, such as the data origin). Both are expected to be of
  type`character` and need to be defined by the backend (i.e., they can not be
  empty or missing).
- `MsBackend` implementations can also represent purely *read-only* data
  resources. In this case only data accessor methods need to be implemented but
  not data replacement methods. Whether a backend is read-only can be set with
  the `@readonly` slot of the virtual `MsBackend` class (the `isReadOnly()`
  function can be used to retrieve the value for this slot). The default is
  `@readonly = FALSE` and thus all data replacement method listed in section
  *Data replacement methods* have to be implemented. For read-only backends
  (`@readonly = TRUE`) only the methods in section *Required methods* need to be
  implemented. Backends can also be *partially* read-only, such as the
  `MsBackendMzR`. This backend allows for example to change spectra variables,
  but not the peaks data (i.e. the m/z and intensity values). Also, backends for
  purely read-only resources could extend the `MsBackendCached` from the
  `r Biocpkg("Spectra")` package to enable support for modifying (or adding)
  spectra variables. Any changes to spectra variables will be internally cached
  by the `MsBackendCached` without the need of them being propagating to the
  underlying data resource (see for example the `MsBackendMassbankSql` from the
  `r Biocpkg("MsBackendMassbank")` package).


## Notes on parallel processing

For parallel processing, `Spectra` splits the backend based on a defined
`factor` and processes each in parallel (or *in serial* if a `SerialParam` is
used). The splitting `factor` can be defined for `Spectra` by setting the
parameter `processingChunkSize`. Alternatively, through the
`backendParallelFactor()` function the backend can also *suggest* a `factor`
that should/could be used for splitting and parallel processing. The default
implementation for `backendParallelFactor()` is to return an empty `factor`
(`factor()`) hence not suggesting any preferred
splitting. `backendParallelFactor()` for `MsBackendMzR` on the other hand
returns a `factor` based on the data files the data is stored in (i.e. based
on the `dataStorage` of the MS data).

Besides parallel processing, this chunk-wise processing can also reduce the
memory demand for operations, because only the peak data of the current chunk
needs to be realized in memory.


# API

The `MsBackend` class defines core methods that have to be implemented by a MS
*backend* as well as *optional* methods with default implementations that might
be implemented for a new backend but don't necessarily have to. These functions
are described in sections *Required methods* and *Optional methods*,
respectively.

To create a new backend a class extending the virtual `MsBackend` needs to be
implemented. In the example below we create thus a simple class with a
`data.frame` to contain general spectral properties (*spectra variables*) and
two slots for m/z and intensity values. These are stored as `NumericList`
objects since both m/z and intensity values are expected to be of type `numeric`
and to allow to store data from multiple spectra into a single backend
object. We also define a simple constructor function that returns an empty
instance of our new class.

```{r, message = FALSE}
library(Spectra)
library(IRanges)

setClass("MsBackendTest",
         contains = "MsBackend",
         slots = c(
             spectraVars = "data.frame",
             mz = "NumericList",
             intensity = "NumericList"
         ),
         prototype = prototype(
             spectraVars = data.frame(),
             mz = NumericList(compress = FALSE),
             intensity = NumericList(compress = FALSE)
         ))

MsBackendTest <- function() {
    new("MsBackendTest")
}
```

The 3 slots `spectraVars`, `mz` and `intensity` will be used to store our MS
data, each row in `spectraVars` being data for one spectrum with the columns
being the different *spectra variables* (i.e. additional properties of a
spectrum such as its retention time or MS level) and each element in `mz` and
`intensity` being a `numeric` with the m/z and intensity values of the
respective spectrum.

We should ideally also add some basic validity function that ensures the data to
be OK. The function below simply checks that the number of rows of the
`spectraVars` slot matches the length of the `mz` and `intensity` slot.

```{r, message = FALSE}
setValidity("MsBackendTest", function(object) {
    if (length(object@mz) != length(object@intensity) ||
        length(object@mz) != nrow(object@spectraVars))
        return("length of 'mz' and 'intensity' has to match the number of ",
               "rows of 'spectraVars'")
    NULL
})
```

We can now create an instance of our new class with the `MsBackendTest()`
function.

```{r}
MsBackendTest()
```

Note that a *backend* class does not necessarily need to contain all the data
like the one from our example. Backends such as the `MsBackendMzR` for example
retrieve the data on the fly from the raw MS data files or the `MsBackendSql`
from the `r Biocpkg("MsBackendSql")` a SQL database.


## Required methods

Methods listed in this section must be implemented for a new class extending
`MsBackend`. Methods should ideally also implemented in the order they are
listed here. Also, it is strongly advised to write dedicated unit tests for
each newly implemented method or function already **during** the development.


### `dataStorage()`

The `dataStorage` spectra variable of a spectrum provides some information how
or where the data is stored. The `dataStorage()` method should therefor return a
`character` vector with length equal to the number of spectra of a backend
object with that information. For most backends the data storage information can
be a simple string such as `"memory"` or `"database"` to specify that the data
of a spectrum is stored within the object itself or in a database,
respectively.

Backend classes that keep only a subset of the MS data in memory and need to
load data from data files upon request will use this spectra variable to store
and keep track of the original data file for each spectrum. An example is the
`MsBackendMzR` backend that retrieves the MS data on-the-fly from the original
data file(s) whenever m/z or intensity values are requested from the
backend. Calling `dataStorage()` on an `MsBackendMzR` returns thus the names
from the originating files.

For our example backend we define a simple `dataStorage()` method that simply
returns the column `"dataStorage"` from the `@svars` (as a `character`).

```{r}
setMethod("dataStorage", "MsBackendTest", function(object) {
    as.character(object@spectraVars$dataStorage)
})
```


### `length()`

`length()` is expected to return a single `integer` with the total number of
spectra that are available through the backend class. For our example backend we
simply return the number of rows of the `data.frame` stored in the
`@spectraVars` slot.

```{r}
setMethod("length", "MsBackendTest", function(x) {
    nrow(x@spectraVars)
})
```


### `backendInitialize()`

The `backendInitialize()` method is expected to be called after creating an
instance of the backend class and should prepare (initialize) the backend which
in most cases means that MS data is loaded. This method can take any parameters
needed by the backend to get loaded/initialized with data (which can be file
names from which to load the data, a database connection or object(s) containing
the data). During `backendInitialize()` usually also the special spectra
variables `dataStorage` and `dataOrigin` are set.

Below we define a `backendInitialize()` method that takes as arguments a
`data.frame` with spectra variables and two `list`s with the m/z and intensity
values for each spectrum.

```{r}
setMethod(
    "backendInitialize", "MsBackendTest",
    function(object, svars, mz, intensity) {
        if (!is.data.frame(svars))
            stop("'svars' needs to be a 'data.frame' with spectra variables")
        if (is.null(svars$dataStorage))
            svars$dataStorage <- "<memory>"
        if (is.null(svars$dataOrigin))
            svars$dataOrigin <- "<user provided>"
        object@spectraVars <- svars
        object@mz <- NumericList(mz, compress = FALSE)
        object@intensity <- NumericList(intensity, compress = FALSE)
        validObject(object)
        object
    })
```

In addition to adding the data to object, the function also defined the
`dataStorage` and `dataOrigin` spectra variables. The purpose of these two
variables is to provide some information on where the data is stored (*in
memory* as in our example) and from where the data is originating. The
`dataOrigin` would for example allow to specify from which original data files
individual spectra derive.

We can now create an instance of our backend class and fill it with data. We
thus first define our MS data and pass this to the `backendInitialize()` method.

```{r}
## A data.frame with spectra variables.
svars <- data.frame(msLevel = c(1L, 2L, 2L),
                    rtime = c(1.2, 1.3, 1.4))
## m/z values for each spectrum.
mzs <- list(c(12.3, 13.5, 16.5, 17.5),
            c(45.1, 45.2),
            c(64.4, 123.1, 124.1))
## intensity values for each spectrum.
ints <- list(c(123.3, 153.6, 2354.3, 243.4),
             c(100, 80.1),
             c(12.3, 35.2, 100))

## Create and initialize the backend
be <- backendInitialize(MsBackendTest(),
                        svars = svars, mz = mzs, intensity = ints)
be
```

While this method works and is compliant with the `MsBackend` API (because there
is no requirement on the input parameters for the `backendInitialize()` method),
it would be good practice for backends that are supposed to support replacing
data, to add an optional additional parameter `data` that would allow passing
the *complete* MS data (including m/z and intensity values) to the function as a
`DataFrame`. This would simplify the implementation of some replacement methods
and would in addition also allow to change the backend of a `Spectra` using the
`setBackend()` to our new backend. We thus re-implement the
`backendInitialize()` method supporting also to initialize the backend with such
a data frame and we also implement a helper function that checks spectra
variables for the correct data type.

```{r}
#' Helper function to check if core spectra variables have the correct
#' data type.
#'
#' @param x `data.frame` with the data for spectra variables.
#'
#' @param name `character` defining the column names (spectra variables) of `x`
#'     for which the correct data type should be evaluated.
.sv_valid_data_type <- function(x, name = colnames(x)) {
    sv <- coreSpectraVariables()[names(coreSpectraVariables()) %in% name]
    for (i in seq_along(sv)) {
        if (!is(x[, names(sv[i])], sv[i]))
            stop("Spectra variabe \"", names(sv[i]), "\" is not of type ",
                 sv[i], call. = FALSE)
    }
    TRUE
}
```

This function is then used to check the input data in our new
`backendInitialize()` method.

```{r}
setMethod(
    "backendInitialize", "MsBackendTest",
    function(object, svars, mz, intensity, data) {
        if (!missing(data)) {
            svars <- as.data.frame(
                data[, !colnames(data) %in% c("mz", "intensity")])
            if (any(colnames(data) == "mz"))
                mz <- data$mz
            if (any(colnames(data) == "intensity"))
                intensity <- data$intensity
        }
        if (!is.data.frame(svars))
            stop("'svars' needs to be a 'data.frame' with spectra variables")
        if (is.null(svars$dataStorage))
            svars$dataStorage <- "<memory>"
        if (is.null(svars$dataOrigin))
            svars$dataOrigin <- "<user provided>"
        .sv_valid_data_type(svars)
        object@spectraVars <- svars
        object@mz <- NumericList(mz, compress = FALSE)
        object@intensity <- NumericList(intensity, compress = FALSE)
        validObject(object)
        object
    })
```

We below create the backend again with the updated `backendInitialize()`.

```{r}
## Create and initialize the backend
be <- backendInitialize(MsBackendTest(),
                        svars = svars, mz = mzs, intensity = ints)
be
```

The `backendInitialize()` method that we implemented for our backend class
expects the user to provide the full MS data. This does however not always have
to be the case. The `backendInitialize()` method of the `MsBackendMzR` backend
takes for example the file names of the raw mzML, mzXML or CDF files as input
and initializes the backend by importing part of the data from these. Also the
backends defined by the `r Biocpkg("MsBackendMgf")` or `r
Biocpkg("MsBackendMsp")` packages work in the same way and thus allow to import
MS data from these specific file formats. The `backendInitialize()` method of
the backend defined in the `r Biocpkg("MsBackendSql")` on the other hand takes
only the connection to a database containing the data as input and performs some
sanity checks on the data but does not load the data into the backend. Any
subsequent data access is handled by the methods of the backend class through
SQL calls to the database.

The purpose of the `backendInitialize()` method is to *initialize* and prepare
the data in a way that it can be accessed by a `Spectra` object (through the
initialized backend class). Whether the data is loaded by the
`backendInitialize()` method into memory or simply referenced to within the
backend class does not matter as long as the backend is able to provide the data
with its accessor methods.

Note also that a `backendInitialize()` function should ideally also perform some
data sanity checks (e.g. whether spectra variables have the correct data type
etc).


### `spectraVariables()`

The `spectraVariables()` method should return a `character` vector with the
names of all available spectra variables of the backend. While a backend class
should support defining and providing their own spectra variables, each
`MsBackend` class **must** provide also the *core spectra variables* (in the
correct data type). Since not all data file formats provide values for all these
spectra variables they can however also be `NA` (with the exception of the
spectra variable `"dataStorage"`).

The `coreSpectraVariables()` function returns the full list of mandatory spectra
variables along with their expected data type.

```{r}
coreSpectraVariables()
```

A typical `spectraVariables()` method for a `MsBackend` class will thus be
implemented similarly to the one for our `MsBackendTest` test backend: it will
return the union of the core spectra variables and the names for all available
spectra variables within the backend object.

```{r}
setMethod("spectraVariables", "MsBackendTest", function(object) {
    union(names(coreSpectraVariables()), colnames(object@spectraVars))
})
spectraVariables(be)
```


### `spectraData()`

The `spectraData()` method should return the **full** spectra data within a
backend as a `DataFrame` object (defined in the `r Biocpkg("S4Vectors")`
package). The second parameter `columns` allows to define the names of the
spectra variables that should be returned in the `DataFrame`. Each row in this
data frame should represent one spectrum, each column a spectra
variable. Columns `"mz"` and `"intensity"` (if requested) have to contain each a
`NumericList` with the m/z and intensity values of the spectra. The `DataFrame`
**must** provide values (even if they are `NA`) for **all** requested spectra
variables of the backend (**including** the core spectra variables).

This is now a first problem for our toy backend class, since we keep the spectra
variable data in a simple `data.frame` without any constraints such as required
columns etc. A simple solution to this (which is also used by all backend
classes in the `Spectra` package) is to *fill* missing spectra variables
on-the-fly into the returned `DataFrame`. We thus define below a simple helper
function that adds columns with missing values (of the correct data type) for
core spectra variables that are not available within the backend to the result.

```{r}
#' @description Add columns with missing core spectra variables.
#'
#' @param x `data.frame` or `DataFrame` with some spectra variables.
#'
#' @param core_vars `character` with core spectra variable names that should
#'     be added to `x` if not already present.
#'
.fill_core_variables <- function(x, core_vars = names(coreSpectraVariables())) {
    fill_vars <- setdiff(core_vars, colnames(x))
    core_type <- coreSpectraVariables()
    n <- nrow(x)
    if (length(fill_vars)) {
        fill <- lapply(fill_vars, function(z) {
            rep(as(NA, core_type[z]), n)
        })
        names(fill) <- fill_vars
        x <- cbind(x, as.data.frame(fill))
    }
    x
}
```

We next implement the `spectraData()` method that uses this helper function to
fill eventually missing core spectra variables. Note also that this function
should return a `DataFrame` even for a single column.

```{r}
setMethod(
    "spectraData", "MsBackendTest",
    function(object, columns = spectraVariables(object)) {
        if (!all(columns %in% spectraVariables(object)))
            stop("Some of the requested spectra variables are not available")
        ## Add m/z and intensity values to the result
        res <- DataFrame(object@spectraVars)
        res$mz <- object@mz
        res$intensity <- object@intensity
        ## Fill with eventually missing core variables
        res <- .fill_core_variables(
            res, intersect(columns, names(coreSpectraVariables())))
        res[, columns, drop = FALSE]
})
```

As an alternative, we could also initialize the `@spectraVars` data frame within
the `backendInitialize()` method adding columns for spectra variables that are
not provided by the user and require that this data frame always contains all
core spectra variables. Extracting spectra data (single spectra variables or the
full data) might thus be more efficient then the on-the-fly initialization with
eventual missing spectra variables, but the backend class would also have a
larger memory footprint because even spectra variables with only missing values
for all spectra need to be stored within the object.

We can now use `spectraData()` to either extract the full spectra data from the
backend, or only the data for selected spectra variables.

```{r}
## Full data
spectraData(be)

## Selected variables
spectraData(be, c("rtime", "mz", "centroided"))

## Only missing core spectra variables
spectraData(be, c("centroided", "polarity"))
```


### `peaksData()`

The `peaksData()` method extracts the MS peaks data from a backend, which
includes the m/z and intensity values of each MS peak of a spectrum. These are
expected to be returned as a `List` of numerical matrices with columns in each
`matrix` being the requested *peaks variables* (with the default being `"mz"`
and `"intensity"`) of one spectrum. Backends must provide at least these two
peaks variables.

Below we implement the `peaksData()` method for our backend. We need to loop
over the `@mz` and `@intensity` slots to merge the m/z and intensity of each
spectrum into a `matrix`. Also, for simplicity reasons, we accept only `c("mz",
"intensity")` for the `columns` parameter. This is the expected default behavior
for a `MsBackend`, but in general the `columns` parameter is thought to allow
the user to specify which peaks variables should be returned in each `matrix`.

```{r}
setMethod(
    "peaksData", "MsBackendTest",
    function(object, columns = c("mz", "intensity")) {
        if (length(columns) != 2 && columns != c("mz", "intensity"))
            stop("'columns' supports only \"mz\" and \"intensity\"")
        mapply(mz = object@mz, intensity = object@intensity,
               FUN = cbind, SIMPLIFY = FALSE, USE.NAMES = FALSE)
    })
```

And with this method we can now extract the peaks data from our backend.

```{r}
peaksData(be)
```

The `peaksData()` method is used in many data analysis functions of the
`Spectra` object to extract the MS data, thus ideally this method should be
implemented in an efficient way. For our backend we need to loop over the lists
of m/z and intensity values which is obviously not ideal. Thus, storing the m/z
and intensity values in separate slots as done in this backend might not be
ideal. The `MsBackendMemory` backend for example stores the MS data already as a
`list` of matrices which results in a more efficient `peaksData()` method (but
comes also with a larger overhead when adding, replacing or checking MS data).

Note also that while a backend needs to provide m/z and intensity values,
additional peak variables would also be supported. The `MsBackendMemory` class
for example allows to store and provide additional peak variables that can then
be added as additional columns to each returned `matrix`. In this case the
default `peaksVariables()` method should also be overwritten to list the
additionally available variables and the `columns` parameter of the
`peaksData()` method should allow selection of these additional peaks variables
(in addition to the required `"mz"` and `"intensity"` variables).


### `[`

The `[` method allows to subset `MsBackend` objects. This operation is expected
to reduce a `MsBackend` object to the selected spectra. The method should
support to subset by indices or logical vectors and should also support
duplicating elements (i.e. when duplicated indices are used) as well as to
subset in arbitrary order. An error should be thrown if indices are out of
bounds, but the method should also support returning an empty backend with
`[integer()]`. Note that the `MsCoreUtils::i2index` function can be used to
check for correct input (and convert the input to an `integer` index).

Below we implement a possible `[` for our test backend class. We ignore the
parameters `j` from the definition of the `[` generic, since we treat our data
to be one-dimensional (with each spectrum being one element).

```{r}
setMethod("[", "MsBackendTest", function(x, i, j, ..., drop = FALSE) {
    i <- MsCoreUtils::i2index(i, length = length(x))
    x@spectraVars <- x@spectraVars[i, ]
    x@mz <- x@mz[i]
    x@intensity <- x@intensity[i]
    x
})
```

We can now subset our backend to the last two spectra.

```{r}
a <- be[2:3]
spectraData(a)
```

Or extracting the second spectrum multiple times.

```{r}
a <- be[c(2, 2, 2)]
spectraData(a)
```


### `backendMerge()`

The `backendMerge()` method merges (combines) `MsBackend` objects (of the same
type!) into a single instance. For our test backend we thus need to combine the
values in the `@spectraVars`, `@mz` and `@intensity` slots. To support also
merging of `data.frame`s with different set of columns we use the
`MsCoreUtils::rbindFill()` function instead of a simple `rbind()` (this function
joins data frames making an union of all available columns).

```{r}
setMethod("backendMerge", "MsBackendTest", function(object, ...) {
    res <- object
    object <- unname(c(object, ...))
    res@mz <- do.call(c, lapply(object, function(z) z@mz))
    res@intensity <- do.call(c, lapply(object, function(z) z@intensity))
    res@spectraVars <- do.call(MsCoreUtils::rbindFill,
                               lapply(object, function(z) z@spectraVars))
    validObject(res)
    res
})
```

Again, this implementation which requires 3 loops might not be the most
efficient - but it allows to merge backends of the type `MsBackendTest`.

```{r}
a <- backendMerge(be, be[2], be)
a
```


### `$`

The `$` method is expected to extract a single spectra variable from a
backend. Parameter `name` should allow to name the spectra variable to
return. Each `MsBackend` **must** support extracting the core spectra variables
with this method (even if no data might be available for that variable). In our
example implementation below we make use of the `spectraData()` method, but more
efficient implementations might be available as well (that would not require to
first subset/create a `DataFrame` with the full data and to then subset that
again). Also, the `$` method should check if the requested spectra variable is
available and should throw an error otherwise.

```{r}
setMethod("$", "MsBackendTest", function(x, name) {
    spectraData(x, columns = name)[, 1L]
})
```

With this we can now extract the MS levels

```{r}
be$msLevel
```

or a core spectra variable that is not available within the backend.

```{r}
be$precursorMz
```

or also the m/z values

```{r}
be$mz
```


### `lengths()`

The `lengths()` method is expected to return an `integer` vector (same length as
the number of spectra in the backend) with the total number of peaks per
spectrum.

For our `MsBackendTest` we can simply use the `lengths()` method on the m/z or
intensity values for that.

```{r}
setMethod("lengths", "MsBackendTest", function(x, use.names = FALSE) {
    lengths(x@mz, use.names = use.names)
})
```

And we can now get the peaks count per spectrum:

```{r}
lengths(be)
```


### `isEmpty()`

The `isEmpty()` method is expected to return for each spectrum the information
whether it is *empty*, i.e. does not contain any MS peaks (and hence m/z or
intensity values). The result of the method has to be a `logical` of length
equal to the number of spectra represented by the backend with `TRUE` indicating
whether a spectrum is empty and `FALSE` otherwise. For our implementation of the
`isEmpty()` method we use the `lenghts()` method defined above that returns the
number of MS peaks per spectrum.

```{r}
setMethod("isEmpty", "MsBackendTest", function(x) {
    lengths(x) == 0L
})
isEmpty(be)
```


### `acquisitionNum()`

Extract the `acquisitionNum` core spectra variable. The method is expected to
return an `integer` vector with the same length as there are spectra represented
by the backend. For our backend we simply re-use the `spectraData()` method.

```{r}
setMethod("acquisitionNum", "MsBackendTest", function(object) {
    spectraData(object, "acquisitionNum")[, 1L]
})
acquisitionNum(be)
```


### `centroided()`

Extract for each spectrum the information whether it contains *centroided*
data. The method is expected to return a `logical` vector with the same length
as there are spectra represented by the backend.

```{r}
setMethod("centroided", "MsBackendTest", function(object) {
    spectraData(object, "centroided")[, 1L]
})
centroided(be)
```


### `collisionEnergy()`

Extract for each spectrum the collision energy applied to generate the fragment
spectrum. The method is expected to return a `numeric` vector with the same
length as there are spectra represented by the backend (with `NA_real_` for
spectra for which this information is not available, such as MS1 spectra).

```{r}
setMethod("collisionEnergy", "MsBackendTest", function(object) {
    spectraData(object, "collisionEnergy")[, 1L]
})
collisionEnergy(be)
```


### `dataOrigin()`

Extract the *data origin* spectra variable for each spectrum. This spectra
variable can be used to store the origin of each spectra. The method is expected
to return a `character` vector of length equal to the number of spectra
represented by the backend.

```{r}
setMethod("dataOrigin", "MsBackendTest", function(object) {
    spectraData(object, "dataOrigin")[, 1L]
})
dataOrigin(be)
```


### `intensity()`

Extract the intensity values for each spectrum in the backend. The result is
expected to be a `NumericList` of length equal to the number of spectra
represented by the backend. For our test backend we can simply return the
`@intensity` slot since the data is already stored within a `NumericList`.

```{r}
setMethod("intensity", "MsBackendTest", function(object) {
    object@intensity
})
intensity(be)
```


### `isolationWindowLowerMz()`

Extract the core spectra variable `isolationWindowLowerMz` from the
backend. This information is usually provided for each spectrum in the raw mzML
files. The method is expected to return a `numeric` vector of length equal to
the number of spectra represented by the backend.

```{r}
setMethod("isolationWindowLowerMz", "MsBackendTest", function(object) {
    spectraData(object, "isolationWindowLowerMz")[, 1L]
})
isolationWindowLowerMz(be)
```


### `isolationWindowTargetMz()`

Extract the core spectra variable `isolationWindowTargetMz` from the
backend. This information is usually provided for each spectrum in the raw mzML
files. The method is expected to return a `numeric` vector of length equal to
the number of spectra represented by the backend.

```{r}
setMethod("isolationWindowTargetMz", "MsBackendTest", function(object) {
    spectraData(object, "isolationWindowTargetMz")[, 1L]
})
isolationWindowTargetMz(be)
```


### `isolationWindowUpperMz()`

Extract the core spectra variable `isolationWindowUpperMz` from the
backend. This information is usually provided for each spectrum in the raw mzML
files. The method is expected to return a `numeric` vector of length equal to
the number of spectra represented by the backend.

```{r}
setMethod("isolationWindowUpperMz", "MsBackendTest", function(object) {
    spectraData(object, "isolationWindowUpperMz")[, 1L]
})
isolationWindowUpperMz(be)
```


### `msLevel()`

Extract the MS level for each spectrum in the backend. This method is expected
to return an `integer` of length equal to the number of spectra represented by
the backend.

```{r}
setMethod("msLevel", "MsBackendTest", function(object) {
    spectraData(object, "msLevel")[, 1L]
})
msLevel(be)
```


### `mz()`

Extract the m/z values for each spectrum in the backend. The result is
expected to be a `NumericList` of length equal to the number of spectra
represented by the backend. Also, the m/z values are expected to be ordered
increasingly for each element (spectrum).

```{r}
setMethod("mz", "MsBackendTest", function(object) {
    object@mz
})
mz(be)
```


### `polarity()`

Extract the `polarity` core spectra variable for each spectrum in the
backend. This method is expected to return an `integer` of length equal to the
number of spectra represented by the backend. Negative and positive polarity are
expected to be encoded by `0L` and `1L`, respectively.

```{r}
setMethod("polarity", "MsBackendTest", function(object) {
    spectraData(object, "polarity")[, 1L]
})
polarity(be)
```


### `precScanNum()`

Extract the acquisition number of the precursor for each spectrum. This method
is expected to return an `integer` of length equal to the number of spectra
represented by the backend. For MS1 spectra (or if the acquisition number of the
precursor is not provided) `NA_integer_` has to be returned.

```{r}
setMethod("precScanNum", "MsBackendTest", function(object) {
    spectraData(object, "precScanNum")[, 1L]
})
precScanNum(be)
```


### `precursorCharge()`

Extract the charge of the precursor for each spectrum. This method
is expected to return an `integer` of length equal to the number of spectra
represented by the backend. For MS1 spectra (or if the charge of the
precursor is not provided) `NA_integer_` has to be returned.

```{r}
setMethod("precursorCharge", "MsBackendTest", function(object) {
    spectraData(object, "precursorCharge")[, 1L]
})
precursorCharge(be)
```


### `precursorIntensity()`

Extract the intensity of the precursor for each spectrum. This method is
expected to return an `numeric` of length equal to the number of spectra
represented by the backend. For MS1 spectra (or if the precursor intensity for a
fragment spectrum is not provided) `NA_real_` has to be returned.

```{r}
setMethod("precursorIntensity", "MsBackendTest", function(object) {
    spectraData(object, "precursorIntensity")[, 1L]
})
precursorIntensity(be)
```


### `precursorMz()`

Extract the precursor m/z for each spectrum. This method is
expected to return an `numeric` of length equal to the number of spectra
represented by the backend. For MS1 spectra (or if the precursor m/z for a
fragment spectrum is not provided) `NA_real_` has to be returned.

```{r}
setMethod("precursorMz", "MsBackendTest", function(object) {
    spectraData(object, "precursorMz")[, 1L]
})
precursorMz(be)
```


### `rtime()`

Extract the retention time of each spectrum. This method is expected to return a
`numeric` of length equal to the number of spectra represented by the backend.

```{r}
setMethod("rtime", "MsBackendTest", function(object) {
    spectraData(object, "rtime")[, 1L]
})
rtime(be)
```


### `scanIndex()`

Extract the *scan index* core spectra variable. The scan index represents the
relative index of the spectrum within the respective raw data file and can be
different than the `acquisitionNum` (which is the index of a spectrum as
recorded by the MS instrument). This method is expected to return a `integer` of
length equal to the number of spectra represented by the backend.

```{r}
setMethod("scanIndex", "MsBackendTest", function(object) {
    spectraData(object, "scanIndex")[, 1L]
})
scanIndex(be)
```


### `smoothed()`

Extract the `smoothed` core spectra variable that indicates whether a spectrum
was *smoothed*. This variable is supported for backward compatibility but
seldomly used. The method is expected to return a `logical` with length equal to
the number of spectra represented by the backend.

```{r}
setMethod("smoothed", "MsBackendTest", function(object) {
    spectraData(object, "smoothed")[, 1L]
})
smoothed(be)
```


### `spectraNames()`

The `spectraNames()` can be used to extract (optional) names (or IDs) for
individual spectra of a backend, or `NULL` if not set. For our test backend we
can use the `rownames` of the `@spectraVars` slot to store spectra names.

```{r}
setMethod("spectraNames", "MsBackendTest", function(object) {
    rownames(object@spectraVars)
})
spectraNames(be)
```

These are all the methods that need to be implemented for a valid *read-only*
`MsBackend` class and running a test on such an object as described in section
*Testing the validity of the backend* should not produce any errors. For
backends that support also data replacement also the methods listed in the next
section need to be implemented.


### `tic()`

The `tic()` method should return the total ion count (i.e. the sum of
intensities) for each spectrum. This information is usually also provided by the
raw MS data files, but can also be calculated on the fly from the data. The
parameter `initial` (which is by default `TRUE`) allows to define whether the
provided *original* tic should be returned (for `initial = TRUE`) or whether the
tic should be calculated on the actual data (`initial = FALSE`). The original
tic values are usually provided by a spectra variable `"totIonCurrent"`. Thus,
for `initial = TRUE`, in our implementation below we return the value of such a
spectra variable it it is avaialble or `NA` if it is not.

```{r}
setMethod("tic", "MsBackendTest", function(object, initial = TRUE) {
    if (initial) {
        if (any(spectraVariables(object) == "totIonCurrent"))
            spectraData(object, "totIonCurrent")[, 1L]
        else rep(NA_real_, length(object))
    } else vapply(intensity(object), sum, numeric(1), na.rm = TRUE)
})
```

We can now either return the original (initial) TIC (which is not available).

```{r}
tic(be)
```

Or calculate the TIC based on the actual intensity values.

```{r}
tic(be, initial = FALSE)
```


## Data replacement methods

As stated in the general description, `MsBackend` implementations can also be
purely *read-only* resources allowing to just access data, but not to replace
the data. Thus, it is not strictly required to implement these methods, but for
a fully functional backend it is suggested (as much as possible). A backend for
a purely read-only MS data resource might even extend the `MsBackendCached`
backend defined in the `Spectra` package that provides a mechanism to cache
(spectra variable) data in a `data.frame` within the object. The
`MsBackendMassbankSql` implemented in the `r Biocpkg("MsBackendMassbank")`
package extends for example this backend and thus allows modifying some spectra
variables without changing the original data in the MassBank SQL database.


### `spectraData<-`

The `spectraData<-` method should allow to replace the data within a
backend. The method should take a `DataFrame` with the full data as input value
and is expected to replace the **full** data within the backend, i.e. all
spectra variables as well as peak data. Also, importantly, the number of spectra
before and after calling the `spectraData<-` method on an object has to be the
same. For our implementation we can make use of the optional parameter `data`
that we added to the `backendInitialize()` method and that allows to fill a
`MsBackendTest` object with the full data.

```{r}
setReplaceMethod("spectraData", "MsBackendTest", function(object, value) {
    if (!inherits(value, "DataFrame"))
        stop("'value' is expected to be a 'DataFrame'")
    if (length(object) && length(object) != nrow(value))
        stop("'value' has to be a 'DataFrame' with ", length(object), " rows")
    object <- backendInitialize(MsBackendTest(), data = value)
    object
})
```

To test this new method we extract the full spectra data, add an additional
column (spectra variable) and replace the data again.

```{r}
d <- spectraData(be)
d$new_col <- c("a", "b", "c")

spectraData(be) <- d
be$new_col
```


### `intensity<-`

The `intensity<-` method should allow to replace the intensity values of all
spectra in a backend. This method is expected to only replace the *values* of
the intensities, but must not change the number of intensities (and hence peaks)
of a spectrum (that could be done with the `peaksData<-` method that allows to
replace intensity **and** m/z values at the same time). The `value` for the
method should ideally be a `NumericList` to ensure that all intensity values are
indeed `numeric`. In addition to the method we implement also a simple helper
function that checks for the correct length of `value`. Each data replacement
method needs to check for that and this function thus reduces code duplication.

```{r}
.match_length <- function(x, y) {
    if (length(x) != length(y))
        stop("Length of 'value' has to match the length of 'object'")
}

setReplaceMethod("intensity", "MsBackendTest", function(object, value) {
    .match_length(object, value)
    if (!(is.list(value) || inherits(value, "NumericList")))
        stop("'value' has to be a list or NumericList")
    if (!all(lengths(value) == lengths(mz(object))))
        stop("lengths of 'value' has to match the number of peaks per spectrum")
    if (!inherits(value, "NumericList"))
        value <- NumericList(value, compress = FALSE)
    object@intensity <- value
    object
})
```

We could now use this method to replace the intensities in our backend with
modified intensities.

```{r}
intensity(be)
intensity(be) <- intensity(be) - 10
intensity(be)
```


### `mz<-`

The `mz<-` method should allow to replace the m/z values of all spectra in a
backend. The implementation can be the same as for the `intensity<-`
method. m/z values within each spectrum need to be increasingly ordered. We thus
also check that this is the case for the provided m/z values. We take here the
advantage that a efficient `is.unsorted()` implementation for `NumericList` is
already available, which is faster than e.g. calling
`vapply(mz(be), is.unsorted, logical(1))`.

```{r}
setReplaceMethod("mz", "MsBackendTest", function(object, value) {
    .match_length(object, value)
    if (!(is.list(value) || inherits(value, "NumericList")))
        stop("'value' has to be a list or NumericList")
    if (!all(lengths(value) == lengths(mz(object))))
        stop("lengths of 'value' has to match the number of peaks per spectrum")
    if (!inherits(value, "NumericList"))
        value <- NumericList(value, compress = FALSE)
    if (any(is.unsorted(value)))
        stop("m/z values need to be increasingly sorted within each spectrum")
    object@mz <- value
    object
})
```


### `peaksData<-`

The `peaksData<-` should allow to replace the peaks data (m/z and intensity
values) of all spectra in a backend. In contrast to the `mz<-` and `intensity<-`
methods this method should also support changing the number of peaks per
spectrum (e.g. due to filtering). Parameter `value` has to be a `list` of
`matrix` objects with columns `"mz"` and `"intensity"`. The length of this list
has to match the number of spectra in the backend. In the implementation for our
backend class we need to loop over this list to extract the m/z and intensity
values and assign them to the `@mz` and `@intensity` slots.

```{r}
setReplaceMethod("peaksData", "MsBackendTest", function(object, value) {
    if (!(is.list(value) || inherits(value, "SimpleList")))
        stop("'value' has to be a list-like object")
    .match_length(object, value)
    object@mz <- NumericList(lapply(value, "[", , "mz"), compress = FALSE)
    object@intensity <- NumericList(lapply(value, "[", , "intensity"),
                                    compress = FALSE)
    validObject(object)
    object
})
```

Using the `peaksData<-` method we can now also for example remove peaks.

```{r}
pd <- peaksData(be)
## Remove the first peak from the first spectrum
pd[[1L]] <- pd[[1L]][-1L, ]

lengths(be)
peaksData(be) <- pd
lengths(be)
```


### `$<-`

The `$<-` method should allow to replace values for spectra variables or also to
add additional spectra variables to the backend. As with all replacement
methods, the `length()` of `value` has to match the number of spectra
represented by the backend.

```{r}
setReplaceMethod("$", "MsBackendTest", function(x, name, value) {
    .match_length(x, value)
    if (name == "mz") {
        mz(x) <- value
    } else if (name == "intensity") {
       intensity(x) <- value
    } else {
        x@spectraVars[[name]] <- value
    }
    .sv_valid_data_type(x@spectraVars, name)
    x
})
```

We can now replace for example existing spectra variables:

```{r}
msLevel(be)
be$msLevel <- c(2L, 1L, 2L)
msLevel(be)
```

Or even add new spectra variables.

```{r}
be$new_var <- c("a", "b", "c")
be$new_var
```

Replacement methods for all *core* spectra variables can be implemented
similarly.


### `selectSpectraVariables()`

The `selectSpectraVariables()` function should allow to reduce the information
within the backend (parameter `object`) to the selected spectra variables
(parameter `spectraVariables`). This is equivalent to a subset by
columns/variables. For core spectra variables, if not specified by parameter
`spectraVariables`, only their values are expected to be removed (since core
spectra variables are expected to be available even if they are not defined
within a backend). The implementation for our backend will remove any columns in
the `@spectraVars` data frame not defined in the `spectraVariables`
parameter. Special care is given to the `"mz"` and `"intensity"` spectra
variables: if they are not selected, the `@mz` and `@intensity` slots are
initialized with empty `NumericList` (of length matching the number of
spectra). Note also that some backends might throw an error if a spectra
variable required for the backend is removed (such as `"dataStorage"` for a
`MsBackendMzR` backend, which is required by the backend to allow retrieval of
m/z and intensity values).

```{r}
setMethod(
    "selectSpectraVariables", "MsBackendTest",
    function(object, spectraVariables = spectraVariables(object)) {
        keep <- colnames(object@spectraVars) %in% spectraVariables
        object@spectraVars <- object@spectraVars[, keep, drop = FALSE]
        if (!any(spectraVariables == "mz"))
            object@mz <- NumericList(vector("list", length(object)),
                                     compress = FALSE)
        if (!any(spectraVariables == "intensity"))
            object@intensity <- NumericList(vector("list", length(object)),
                                            compress = FALSE)
        validObject(object)
        object
    })
```

We can now use `selectSpectraVariables()` to remove for example the spectra
variable `"new_var"` added above.

```{r}
be2 <- be
be2 <- selectSpectraVariables(be2, c("msLevel", "rtime", "mz",
                                     "intensity", "dataStorage"))
spectraVariables(be2)
```

The spectra variable `"new_var"` is now no longer be available. Note however
that still **all** core spectra variables are listed, even if they were not
selected with the `spectraVariables` parameter. While these variables (such as
`"dataOrigin"`) are still listed by `spectraVariables(be2)`, their actual values
have been removed:

```{r}
dataOrigin(be)
dataOrigin(be2)
```

If `"mz"` and `"intensitity"` are not selected, the m/z and intensity values get
removed.

```{r}
be2 <- selectSpectraVariables(be2, c("msLevel", "rtime", "dataStorage"))
mz(be2)
intensity(be2)
```


### `centroided<-`

Replace the value for the *centroided* core spectra variable. The provided data
type **must** be a logical. We re-use the function `.sv_valid_data_type` which
was defined above for `backendInitialize()` to check for the correct data type
of core spectra variables.

```{r}
setReplaceMethod("centroided", "MsBackendTest", function(object, value) {
    object@spectraVars[["centroided"]] <- value
    .sv_valid_data_type(object@spectraVars, "centroided")
    object
})
```

Alternatively, we could also simply re-use the `$<-` replacement method
above. This would make the whole code base for our backend cleaner as replacing
or adding spectra variables would be handled in a single central function.

```{r}
setReplaceMethod("centroided", "MsBackendTest", function(object, value) {
    object$centroided <- value
    object
})
```

```{r}
centroided(be) <- c(TRUE, FALSE, TRUE)
centroided(be)
```


### `collisionEnergy<-`

Replace the values for the collision energy. Parameter `value` has to be of type
`numeric`.

```{r}
setReplaceMethod("collisionEnergy", "MsBackendTest", function(object, value) {
    object$collisionEnergy <- value
    object
})
```

```{r}
collisionEnergy(be) <- c(NA_real_, 20.0, 20.0)
collisionEnergy(be)
```


### `dataOrigin<-`

Replace the values for the data origin spectra variable. Parameter `value` has
to be of type `character`.

```{r}
setReplaceMethod("dataOrigin", "MsBackendTest", function(object, value) {
    object$dataOrigin <- value
    object
})
```

```{r}
dataOrigin(be)
dataOrigin(be) <- c("unknown", "file a", "file b")
dataOrigin(be)
```


### `dataStorage<-`

Replace the values for the data storage spectra variable. Parameter `value` has
to be of type `character`. Since our backend does not really make any use of
this spectra variable, we can accept any character value. For other backends,
that for example need to load data on-the-fly from data files, this spectra
variable could be used to store the name of the data files and hence we would
need to perform some additional checks within this replacement function.

```{r}
setReplaceMethod("dataStorage", "MsBackendTest", function(object, value) {
    object$dataStorage <- value
    object
})
```

```{r}
dataStorage(be)
dataStorage(be) <- c("", "", "")
dataStorage(be)
```


### `isolationWindowLowerMz<-`

Replace the values for the *isolation window lower m/z* spectra
variable. Parameter `value` has to be of type `numeric` (`NA_real_` missing
values are supported, e.g. for MS1 spectra).

```{r}
setReplaceMethod(
    "isolationWindowLowerMz", "MsBackendTest", function(object, value) {
        object$isolationWindowLowerMz <- value
        object
    })
```

```{r}
isolationWindowLowerMz(be) <- c(NA_real_, 245.3, NA_real_)
isolationWindowLowerMz(be)
```


### `isolationWindowTargetMz<-`

Replace the values for the *isolation window target m/z* spectra
variable. Parameter `value` has to be of type `numeric` (`NA_real_` missing
values are supported, e.g. for MS1 spectra).

```{r}
setReplaceMethod(
    "isolationWindowTargetMz", "MsBackendTest", function(object, value) {
        object$isolationWindowTargetMz <- value
        object
    })
```

```{r}
isolationWindowTargetMz(be) <- c(NA_real_, 245.4, NA_real_)
isolationWindowTargetMz(be)
```


### `isolationWindowUpperMz<-`

Replace the values for the *isolation window upper m/z* spectra
variable. Parameter `value` has to be of type `numeric` (`NA_real_` missing
values are supported, e.g. for MS1 spectra).

```{r}
setReplaceMethod(
    "isolationWindowUpperMz", "MsBackendTest", function(object, value) {
        object$isolationWindowUpperMz <- value
        object
    })
```

```{r}
isolationWindowUpperMz(be) <- c(NA_real_, 245.5, NA_real_)
isolationWindowUpperMz(be)
```


### `msLevel<-`

Replace the MS level of spectra in a backend. Parameter `value` has to be of
type `integer`. Missing values (`NA_integer_`) are supported.

```{r}
setReplaceMethod("msLevel", "MsBackendTest", function(object, value) {
    object$msLevel <- value
    object
})
```

```{r}
msLevel(be)
msLevel(be) <- c(1L, 1L, 2L)
msLevel(be)
```


### `polarity<-`

Replace the values for the polarity spectra variables. Parameter `value` has to
be of type `integer` and should ideally also use the *standard* encoding `0L`
and `1L` for negative and positive polarity (and `NA_integer` for
missing). Thus, in our implementation we also make sure the input parameter
contains the expected values (although this is not a strictly required).

```{r}
setReplaceMethod("polarity", "MsBackendTest", function(object, value) {
    if (!all(value %in% c(0, 1, NA)))
        stop("'polarity' should be encoded as 0L (negative), 1L (positive) ",
             "with missing values being NA_integer_")
    object$polarity <- value
    object
})
```

```{r}
polarity(be) <- c(0L, 0L, 0L)
polarity(be)
```


### `rtime<-`

Replace the retention times for the spectra represented by the
backend. Parameter `value` must be of type `numeric`. Also, although it is not a
strict requirement, retention times should ideally be ordered increasingly *per
sample* and their unit should be seconds.

```{r}
setReplaceMethod("rtime", "MsBackendTest", function(object, value) {
    object$rtime <- value
    object
})
```

```{r}
rtime(be)
rtime(be) <- rtime(be) + 2
rtime(be)
```


### `smoothed<-`

Replace the spectra variable *smoothed* that indicates whether some data
smoothing operation was performed on the spectra. Parameter `value` must be of
type `logical`.

```{r}
setReplaceMethod("smoothed", "MsBackendTest", function(object, value) {
    object$smoothed <- value
    object
})
```

```{r}
smoothed(be) <- rep(TRUE, 3)
smoothed(be)
```


### `spectraNames<-`

Replace the names of individual spectras within the backend. Same as for
`names`, `colnames` or `rownames`, `spectraNames` are expected to be of type
`character`. In our backend implementation we store the spectra names into the
`rownames` of the `@spectraVars` data frame.

```{r}
setReplaceMethod("spectraNames", "MsBackendTest", function(object, value) {
    rownames(object@spectraVars) <- value
    object
})
```

```{r}
spectraNames(be) <- c("a", "b", "c")
spectraNames(be)
```


## Optional methods

Default implementations for these methods are available for `MsBackend`
classes, thus these methods don't have to be implemented for each new
backend. For some backends, depending on how the data is represented or accessed
within it, different implementations might however be more efficient.

### `backendBpparam()`

The `backendBpparam()` method is supposed to evaluate whether a provided (or the
default) parallel processing setup is supported by the backend. Backends that
do not support parallel processing should return `SerialParam()` instead.

The default implementation is shown below.

```{r}
setMethod("backendBpparam", signature = "MsBackend",
          function(object, BPPARAM = bpparam()) {
              ## Return SerialParam() instead to disable parallel processing
              BPPARAM
          })

```

### `backendParallelFactor()`

The `backendParallelFactor()` allows a backend to suggest a preferred way how
the backend could be split for parallel processing. See also the notes on
parallel processing above for more information. The default implementation
returns `factor()` (i.e. a `factor` of length 0) hence not suggesting any
splitting:

```{r}
setMethod("backendParallelFactor", "MsBackend", function(object, ...) {
    factor()
})

```

### `dropNaSpectraVariables()`

The `dropNaSpectraVariables()` is supposed to allow removing all spectra
variables from a data set (storage) that contain only missing values (i.e. where
the value of a spectra variable for each spectrum is `NA`). This function is
intended to reduce memory requirements of backends such as the `MsBackendMzR`
that load values from all core spectra variables from the original data files,
even if their values are only `NA`. Removing these missing values from the
backend can hence reduce the size in memory of a backend without data loss
(because methods extracting core spectra variables are supposed to always return
`NA` values even if no data is available for them - in such cases the `NA`
values are supposed to be created on-the-fly.

The default implementation is shown below.

```{r}
setMethod("dropNaSpectraVariables", "MsBackend", function(object) {
    svs <- spectraVariables(object)
    svs <- svs[!(svs %in% c("mz", "intensity"))]
    spd <- spectraData(object, columns = svs)
    keep <- !vapply1l(spd, function(z) all(is.na(z)))
    selectSpectraVariables(object, c(svs[keep], "mz", "intensity"))
})
```

### `isReadOnly()`

`isReadOnly()` is expected to return a `logical(1)` with either `TRUE` or
`FALSE` indicating whether the backend supports replacing data or not. The
default implementation is shown below.

```{r}
setMethod("isReadOnly", "MsBackend", function(object) {
    object@readonly
})
```


### `peaksVariables()`

The `peaksVariables()` is expected to return a `character` vector with the names
of the *peaks variables* (i.e. information and properties of individual mass
peaks) available in the backend. The default implementation for `MsBackend`
returns by default `c("mz", "intensity")`. This method should only be
implemented for backends that (eventually) also provide additional peaks
variables. The default implementation is shown below.

```{r}
setMethod("peaksVariables", "MsBackend", function(object) {
    c("mz", "intensity")
})
```


### `uniqueMsLevels()`

This method should return the unique MS level(s) of all spectra within the
backend. The default implementation is shown below.

```{r}
setMethod("uniqueMsLevels", "MsBackend", function(object, ...) {
    unique(msLevel(object))
})
```

This method thus retrieves first the MS levels of all spectra and then calls
`unique()` on them. Database-based backends might avoid such an eventually heavy
operation by selecting the unique MS levels directly using an SQL call.


### `ionCount()`

The `ionCount()` method should return a `numeric` (length equal to the number of
spectra represented by the backend) with the sum of all intensities within each
spectrum. For empty spectra `NA_real_` should be returned. The method below is
the default implementation of the method.

```{r}
setMethod("ionCount", "MsBackend", function(object) {
    vapply1d(intensity(object), sum, na.rm = TRUE)
})
```


### `isCentroided()`

This method should return the information for each spectrum whether it is
*centroided*. In contrast to the `centroided()` method a heuristic approach is
used. The default implementation is shown below.

```{r}
setMethod("isCentroided", "MsBackend", function(object, ...) {
    vapply1l(peaksData(object), .peaks_is_centroided)
})
```


### `reset()`

This is a special method that backends may implement or support, but don't
necessary have to. This method will be called by the `reset,Spectra` method and
is supposed to restore the data to its original state. The default
implementation for `MsBackend` shown below simply returns the backend as-is. The
`MsBackendSql` backend from the `r Biocpkg("MsBackendSql")` package in contrast
re-initializes the data using the data from the database.

```{r}
setMethod("reset", "MsBackend", function(object) {
    object
})
```


### `export()`

This method should *export* the data from a `MsBackend`. The method is called by
the `export,Spectra` method that passes itself as a second argument to the
function. The `export,MsBackend` implementation is thus expected to take a
`Spectra` object as second argument from which all data should be taken and
exported. Implementation of this method is optional. The implementation of the
method for the `MsBackendMzR` backend is shown below.

```{r, eval = FALSE}
setMethod("export", "MsBackendMzR", function(object, x, file = tempfile(),
                                             format = c("mzML", "mzXML"),
                                             copy = FALSE,
                                             BPPARAM = bpparam()) {
    l <- length(x)
    file <- sanitize_file_name(file)
    if (length(file) == 1)
        file <- rep_len(file, l)
    if (length(file) != l)
        stop("Parameter 'file' has to be either of length 1 or ",
             length(x), ", i.e. 'length(x)'.", call. = FALSE)
    f <- factor(file, levels = unique(file))
    tmp <- bpmapply(.write_ms_data_mzR, split(x, f), levels(f),
                    MoreArgs = list(format = format, copy = copy),
                    BPPARAM = BPPARAM)
})
```

See alternatively also the `r Biocpkg("MsBackendMgf")` package for an
implementation for the `MsBackendMgf` backend.


### `split()`

The `split()` method should allow to split a `MsBackend` into a `list` of
`MsBackend` objects. The default implementation is shown below.

```{r}
setMethod("split", "MsBackend", function(x, f, drop = FALSE, ...) {
    split.default(x, f, drop = drop, ...)
})
```


### `supportsSetBackend()`

Whether a `MsBackend` supports the `setBackend()` method that allows to change
the backend of a `Spectra` object from one to another backend. To support
`setBackend()` the backend needs to have a parameter `data` in its
`backendInitialize()` method that allows it to be initialized with a `DataFrame`
containing the full spectra data. The default implementation is shown below.

```{r}
setMethod("supportsSetBackend", "MsBackend", function(object, ...) {
    !isReadOnly(object)
})
```


### `filterDataOrigin()`

The `filter*` methods are expected to take a `MsBackend` and to subset it based
on some criteria. While also the `[` method could be used to perform such subset
operation, these methods might allow more efficient ways to subset the data
e.g. by performing the operation within a database with a dedicated SQL call. A
default implementation is available for every filter function and thus a method
needs only to be implemented if the data storage/representation within a backend
would allow a more efficient operation.

All filter methods are expected to return the subset backend (i.e. an instance
of the same backend class with the same, or less spectra).

The `filterDataOrigin()` should subset the backend to spectra with their
`dataOrigin` spectra variable matching the values provided with the `dataOrigin`
parameter. The default implementation for `MsBackend` is shown below.

```{r}
setMethod("filterDataOrigin", "MsBackend",
          function(object, dataOrigin = character()) {
              if (length(dataOrigin)) {
                  object <- object[dataOrigin(object) %in% dataOrigin]
                  if (is.unsorted(dataOrigin))
                      object[order(match(dataOrigin(object), dataOrigin))]
                  else object
              } else object
          })

```


### `filterDataStorage()`

Similar to the `filterDataOrigin()`, the `filterDataStorage()` should subset a
backend to spectra with their `dataStorage` spectra variable matching the values
provided with the `dataStorage` parameter.

```{r}
setMethod("filterDataStorage", "MsBackend",
          function(object, dataStorage = character()) {
              if (length(dataStorage)) {
                  object <- object[dataStorage(object) %in% dataStorage]
                  if (is.unsorted(dataStorage))
                      object[order(match(dataStorage(object), dataStorage))]
                  else object
              } else object
          })
```


### `filterEmptySpectra()`

The `filterEmptySpectra()` should remove all *empty* spectra (i.e. spectra
without any mass peaks) from the backend. The method is expected to return the
subset backend. The default implementation for `MsBackend` is shown below.

```{r}
setMethod("filterEmptySpectra", "MsBackend", function(object, ...) {
    if (!length(object)) return(object)
    object[as.logical(lengths(object))]
})
```


### `filterIsolationWindow()`

The `filterIsolationWindow()` filters the backend to spectra with the provided
`mz` value being within their `isolationWindowLowerMz()` and
`isolationWindowUpperMz()`. The parameter `mz` defining this target m/z is
expected to be a `numeric` of length 1. The default implementation for
`MsBackend` is shown below.

```{r}
setMethod("filterIsolationWindow", "MsBackend",
          function(object, mz = numeric(), ...) {
              if (length(mz)) {
                  if (length(mz) > 1)
                      stop("'mz' is expected to be a single m/z value")
                  keep <- which(isolationWindowLowerMz(object) <= mz &
                                isolationWindowUpperMz(object) >= mz)
                  object[keep]
              } else object
          })
```


### `filterMsLevel()`

The `filterMsLevel()` method is expected to reduce the backend to spectra with
the provided MS level(s). Parameter `msLevel` has to be an `integer` (any
length).  The default implementation for `MsBackend` is shown below.

```{r}
setMethod("filterMsLevel", "MsBackend",
          function(object, msLevel = integer()) {
              if (length(msLevel)) {
                  object[msLevel(object) %in% msLevel]
              } else object
          })
```


### `filterPolarity()`

The `filterPolarity()` method is expected to subset the backend to spectra
matching the provided polarity (or polarities). Parameter `polarity` has to be
an `integer` (of any length). The default implementation for `MsBackend` is
shown below.

```{r}
setMethod("filterPolarity", "MsBackend",
          function(object, polarity = integer()) {
              if (length(polarity))
                  object[polarity(object) %in% polarity]
              else object
          })
```


### `filterPrecursorMzRange()`

The `filterPrecursorMzRange()` method filters the backend to spectra with their
`precursorMz` being between the provided m/z range (parameter `mz`). This method
was previously named `filterPrecursorMz()`. Parameter `mz` is expected to be a
`numeric` of length 2 defining the lower and upper limit of this precursor m/z
range. The default implementation for `MsBackend` is shown below.

```{r}
library(MsCoreUtils)
setMethod("filterPrecursorMzRange", "MsBackend",
          function(object, mz = numeric()) {
              if (length(mz)) {
                  mz <- range(mz)
                  keep <- which(between(precursorMz(object), mz))
                  object[keep]
              } else object
          })
```


### `filterPrecursorMzValues()`

The `filterPrecursorMzValues()` method filters the backend to spectra with their
m/z values matching to the provided m/z value(s). Parameters `ppm` and
`tolerance` (both expected to be `numeric` of length 1) allow to define the
conditions for the relaxed matching. Parameter `mz` has to be a `numeric` (of
any length). The default implementation for `MsBackend` is shown below.

```{r}
setMethod("filterPrecursorMzValues", "MsBackend",
          function(object, mz = numeric(), ppm = 20, tolerance = 0) {
              if (length(mz)) {
                  object[.values_match_mz(precursorMz(object), mz = mz,
                                          ppm = ppm, tolerance = tolerance)]
              } else object
          })
```

The `.values_match_mz` function used by this function is defined as:

```{r}
.values_match_mz <- function(x, mz, ppm = 20, tolerance = 0) {
    o <- order(x, na.last = NA)
    cmn <- common(x[o], sort(mz), tolerance = tolerance, ppm = ppm,
                  duplicates = "keep", .check = FALSE)
    sort(o[cmn])
}
```


### `filterPrecursorCharge()`

The `filterPrecursorCharge()` method filters the backend to spectra with
matching precursor charge. Parameter `z` defining the requested precursor charge
has to be an `integer` (of any length). The default implementation for
`MsBackend` is shown below.

```{r}
setMethod("filterPrecursorCharge", "MsBackend",
          function(object, z = integer()) {
              if (length(z)) {
                  keep <- which(precursorCharge(object) %in% z)
                  object[keep]
              } else object
          })
```


### `filterPrecursorScan()`

The `filterPrecursorScan()` method filters the backend to parent (e.g. MS1) and
children scans (e.g. MS2) of acquisition number `acquisitionNum`. Parameter `f`
defines how the backend should be split (by default by original data file) to
avoid selecting spectra from different samples/files. The default implementation
for `MsBackend` is shown below.

```{r}
setMethod("filterPrecursorScan", "MsBackend",
          function(object, acquisitionNum = integer(), f = dataOrigin(object)) {
              if (length(acquisitionNum) && length(f)) {
                  if (!is.factor(f))
                      f <- factor(f, exclude = character())
                  keep <- unsplit(lapply(split(object, f = f), function(z, an) {
                      .filterSpectraHierarchy(acquisitionNum(z),
                                              precScanNum(z),
                                              an)
                  }, an = acquisitionNum), f = f)
                  object[keep]
              } else object
          })
```


### `filterRt()`

The `filterRt()` method filters the backend to spectra with their retention time
being between the provided rt range. Parameter `rt` is expected
to be a `numeric` of length 2 defining the lower and upper bound of this
range. Parameter `msLevel.` (note the `.` in the name of the parameter!) can be
optionally used to restrict the filtering to the selected MS levels (i.e. the RT
filter is only applied to spectra of the selected MS levels and all spectra with
a different MS level are returned as well). The default implementation for
`MsBackend` is shown below.

```{r}
setMethod("filterRt", "MsBackend",
          function(object, rt = numeric(), msLevel. = uniqueMsLevels(object)) {
              if (length(rt)) {
                  rt <- range(rt)
                  sel_ms <- msLevel(object) %in% msLevel.
                  sel_rt <- between(rtime(object), rt) & sel_ms
                  object[sel_rt | !sel_ms]
              } else object
          })
```


## Implementation notes

In this tutorial we implemented a simple *in-memory* `MsBackend` from
scratch. For many real-life situation it might however be better to extend some
of the pre-defined backend classes from the `Spectra` package to avoid
duplicating functionality. A good starting point might be the `MsBackendMemory`
backend for any *in-memory* data representation, or the `MsBackendCached` for
backends that retrieve data from inherently read-only resources (such as
database connection or raw data files) but still would need to support adding
spectra variables or changing values of spectra variables. Similarly, if the
only purpose of a backend is to import or export data in a specific format, the
`MsBackendMemory` might be extended and a single method (`backendInitialize()`)
would need to be implemented for the new class: this new `backendInitialize()`
would then call the code to import the data from the new file format and store
it within the available slots of the `MsBackendMemory` object. Examples would be
the backends provided by the `r Biocpkg("MsBackendMgf")`
and `r Biocpkg("MsBackendMsp")` classes.


# Testing the validity of the backend

The `Spectra` package provides a set of unit tests that allow to check a backend
for compliance with `MsBackend`. Below we load this test suite and call the
tests. The tests will be performed on a variable `be` in the current workspace
(which in our case is an instance of our `MsBackendTest` class).

```{r, eval = FALSE}
library(testthat)
test_suite <- system.file("test_backends", "test_MsBackend",
                          package = "Spectra")
test_dir(test_suite, stop_on_failure = TRUE)
```


# Session information

```{r si}
sessionInfo()
```

# References