---
title: "Description and usage of Spectra objects"
output:
    BiocStyle::html_document:
        toc_float: true
vignette: >
    %\VignetteIndexEntry{Description and usage of Spectra object}
    %\VignetteEngine{knitr::rmarkdown}
    %\VignetteEncoding{UTF-8}
    %\VignettePackage{Spectra}
    %\VignetteDepends{Spectra,mzR,BiocStyle,msdata,msentropy}
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)
register(SerialParam())
```

# Introduction

The `r Biocpkg("Spectra")` package provides a scalable and flexible
infrastructure to represent, retrieve and handle mass spectrometry (MS)
data. The `Spectra` object provides the user with a single standardized
interface to access and manipulate MS data while supporting, through the concept
of exchangeable *backends*, a large variety of different ways to store and
retrieve mass spectrometry data. Such backends range from mzML/mzXML/CDF files,
simple flat files, or database systems.

This vignette provides general examples and descriptions for the *Spectra*
package. Additional information and tutorials are available, such as
[SpectraTutorials](https://jorainer.github.io/SpectraTutorials/),
[MetaboAnnotationTutorials](https://jorainer.github.io/MetaboAnnotationTutorials),
or also in [@rainer_modular_2022]. For information on how to handle and
(parallel) process large-scale data sets see the *Large-scale data handling and
processing with Spectra* vignette.


# Installation

The package can be installed with the *BiocManager* package. To
install *BiocManager* use `install.packages("BiocManager")` and, after that,
`BiocManager::install("Spectra")` to install *Spectra*.


# General usage

Mass spectrometry data in `Spectra` objects can be thought of as a list of
individual spectra, with each spectrum having a set of variables associated with
it. Besides *core* spectra variables (such as MS level or retention time)
an arbitrary number of optional variables can be assigned to a spectrum. The
core spectra variables all have their own accessor method and it is
guaranteed that a value is returned by it (or `NA` if the information is not
available). The core variables and their data type are (alphabetically
ordered):

- *acquisitionNum* `integer(1)`: the index of acquisition of a spectrum during a
  MS run.
- *centroided* `logical(1)`: whether the spectrum is in profile or centroid
  mode.
- *collisionEnergy* `numeric(1)`: collision energy used to create an MSn
  spectrum.
- *dataOrigin* `character(1)`: the *origin* of the spectrum's data, e.g. the
  mzML file from which it was read.
- *dataStorage* `character(1)`: the (current) storage location of the spectrum
  data. This value depends on the backend used to handle and provide the
  data. For an *in-memory* backend like the `MsBackendDataFrame` this will be
  `"<memory>"`, for an on-disk backend such as the `MsBackendHdf5Peaks` it will
  be the name of the HDF5 file where the spectrum's peak data is stored.
- *intensity* `numeric`: intensity values for the spectrum's peaks.
- *isolationWindowLowerMz* `numeric(1)`: lower m/z for the isolation window in
  which the (MSn) spectrum was measured.
- *isolationWindowTargetMz* `numeric(1)`: the target m/z for the isolation
  window in which the (MSn) spectrum was measured.
- *isolationWindowUpperMz* `numeric(1)`: upper m/z for the isolation window in
  which the (MSn) spectrum was measured.
- *msLevel* `integer(1)`: the MS level of the spectrum.
- *mz* `numeric`: the m/z values for the spectrum's peaks.
- *polarity* `integer(1)`: the polarity of the spectrum (`0` and `1`
  representing negative and positive polarity, respectively).
- *precScanNum* `integer(1)`: the scan (acquisition) number of the precursor for
  an MSn spectrum.
- *precursorCharge* `integer(1)`: the charge of the precursor of an MSn
  spectrum.
- *precursorIntensity* `numeric(1)`: the intensity of the precursor of an MSn
  spectrum.
- *precursorMz* `numeric(1)`: the m/z of the precursor of an MSn spectrum.
- *rtime* `numeric(1)`: the retention time of a spectrum.
- *scanIndex* `integer(1)`: the index of a spectrum within a (raw) file.
- *smoothed* `logical(1)`: whether the spectrum was smoothed.

For details on the individual variables and their getter/setter function see the
help for `Spectra` (`?Spectra`). Also note that these variables are suggested,
but not required to characterize a spectrum. Also, some only make sense for MSn,
but not for MS1 spectra.


## Creating `Spectra` objects

The simplest way to create a `Spectra` object is by defining a `DataFrame` with
the corresponding spectra data (using the corresponding spectra variable names
as column names) and passing that to the `Spectra` constructor function. Below
we create such an object for a set of 3 spectra providing their MS level,
olarity but also additional annotations such as their ID in
[HMDB](http://hmdb.ca) (human metabolome database) and their name. The m/z and
intensity values for each spectrum have to be provided as a `list` of `numeric`
values.

```{r spectra-dataframe, message = FALSE}
library(Spectra)

spd <- DataFrame(
    msLevel = c(2L, 2L, 2L),
    polarity = c(1L, 1L, 1L),
    id = c("HMDB0000001", "HMDB0000001", "HMDB0001847"),
    name = c("1-Methylhistidine", "1-Methylhistidine", "Caffeine"))

## Assign m/z and intensity values.
spd$mz <- list(
    c(109.2, 124.2, 124.5, 170.16, 170.52),
    c(83.1, 96.12, 97.14, 109.14, 124.08, 125.1, 170.16),
    c(56.0494, 69.0447, 83.0603, 109.0395, 110.0712,
      111.0551, 123.0429, 138.0662, 195.0876))
spd$intensity <- list(
    c(3.407, 47.494, 3.094, 100.0, 13.240),
    c(6.685, 4.381, 3.022, 16.708, 100.0, 4.565, 40.643),
    c(0.459, 2.585, 2.446, 0.508, 8.968, 0.524, 0.974, 100.0, 40.994))

sps <- Spectra(spd)
sps
```

Alternatively, it is possible to import spectra data from mass spectrometry raw
files in mzML/mzXML or CDF format. Below we create a `Spectra` object from two
mzML files and define to use a `MsBackendMzR` backend to *store* the data (note
that this requires the `r Biocpkg("mzR")` package to be installed). This
backend, specifically designed for raw MS data, keeps only a subset of spectra
variables in memory while reading the m/z and intensity values from the original
data files only on demand. See section [Backends](#backends) for more details
on backends and their properties.

```{r spectra-msbackendmzr, message = FALSE}
fls <- dir(system.file("sciex", package = "msdata"), full.names = TRUE)
sps_sciex <- Spectra(fls, source = MsBackendMzR())
sps_sciex
```

The `Spectra` object `sps_sciex` allows now to access spectra data from 1862 MS1
spectra and uses `MsBackendMzR` as backend (the `Spectra` object `sps` created
in the previous code block uses the default `MsBackendMemory`).


## Accessing spectrum data

As detailed above `Spectra` objects can contain an arbitrary number of
properties of a spectrum (so called *spectra variables*). The available
variables can be listed with the `spectraVariables()` method:

```{r spectravariables}
spectraVariables(sps)
spectraVariables(sps_sciex)
```

The two `Spectra` contain a different set of variables: besides `"msLevel"`,
`"polarity"`, `"id"` and `"name"`, that were specified for the `Spectra` object
`sps`, it contains more variables such as `"rtime"`, `"acquisitionNum"` and
`"scanIndex"`. These are part of the *core variables* defining a spectrum and
for all of these accessor methods exist. Below we use `msLevel()` and `rtime()`
to access the MS levels and retention times for the spectra in `sps`.

```{r mslevel-sps}
msLevel(sps)
rtime(sps)
```

We did not specify retention times for the spectra in `sps` thus `NA` is
returned for them. The `Spectra` object `sps_sciex` contains many more
variables, all of which were extracted from the mzML files. Below we extract the
retention times for the first spectra in the object.

```{r rtime-spssciex}
head(rtime(sps_sciex))
```

Note that in addition to the accessor functions it is also possible to use `$`
to extract a specific spectra variable. To extract the name of the compounds in
`sps` we can use `sps$name`, or, to extract the MS levels `sps$msLevel`.

```{r dollar-extract}
sps$name
sps$msLevel
```

We could also replace specific spectra variables using either the dedicated
method or `$`. Below we specify that all spectra in `sps` represent centroided
data.

```{r dollar-set}
sps$centroided <- TRUE

centroided(sps)
```

The `$` operator can also be used to add arbitrary new spectra variables to a
`Spectra` object. Below we add the SPLASH key to each of the spectra.

```{r new-spectra-variable}
sps$splash <- c(
    "splash10-00di-0900000000-037d24a7d65676b7e356",
    "splash10-00di-0900000000-03e99316bd6c098f5d11",
    "splash10-000i-0900000000-9af60e39c843cb715435")
```

This new spectra variable will now be listed as an additional variable in the
result of the `spectraVariables()` function and we can directly access its
content with `sps$splash`.

Each spectrum can have a different number of mass peaks, each consisting of a
mass-to-charge (m/z) and associated intensity value. These can be extracted with
the `mz()` or `intensity()` functions, each of which return a `list` of
`numeric` values.

```{r mz-intensity}
mz(sps)
intensity(sps)
```

Peak data can also be extracted with the `peaksData()` function that returns a
list of numerical matrices with *peak variables* such as m/z and intensity
values. Which peak variables are available in a `Spectra` object can be
determined with the `peaksVariables()` function.

```{r}
peaksVariables(sps)
```

These can be passed to the `peaksData()` function with parameter `columns` to
extract the peak variables of interest. By default `peaksData()` extracts m/z
and intensity values.

```{r peaks}
pks <- peaksData(sps)
pks[[1]]
```

Note that we would get the same result by using the `as()` method to coerce a
`Spectra` object to a `list` or `SimpleList`:

```{r as}
as(sps, "SimpleList")
```

The `spectraData()` function returns a `DataFrame` with the full data for each
spectrum (except m/z and intensity values), or with selected spectra variables
(which can be specified with the `columns` parameter). Below we extract the
spectra data for variables `"msLevel"`, `"id"` and `"name"`.

```{r spectradata}
spectraData(sps, columns = c("msLevel", "id", "name"))
```

`Spectra` are one-dimensional objects storing spectra, even from different files
or samples, in a single list. Specific variables have thus to be used to define
the originating file from which they were extracted or the sample in which they
were measured. The *data origin* of each spectrum can be extracted with the
`dataOrigin()` function. For `sps`, the `Spectra` created from a `DataFrame`,
this will be `NA` because we did not specify the data origin:

```{r dataOrigin-sps}
dataOrigin(sps)
```

`dataOrigin` for `sps_sciex`, the `Spectra` which was initialized with data
from mzML files, in contrast, returns the originating file names:

```{r dataOrigin-sciex}
head(basename(dataOrigin(sps_sciex)))
```

The current data storage location of a spectrum can be retrieved with the
`dataStorage` variable, which will return an arbitrary string for `Spectra` that
use an in-memory backend or the file where the data is stored for on-disk
backends:

```{r dataStorage}
dataStorage(sps)
head(basename(dataStorage(sps_sciex)))
```

Certain backends (such as the `MsBackendMemory` and `MsBackendDataFrame`)
support also additional peaks variables. At present, these must already be
present when the backend gets initialized. In future a dedicated function
allowing to add peaks variables will be available. Below we thus first extract
the full data (including peaks variables) from the `sps` spectra object and add
a column `"peak_anno"` with *peak annotations* for each individual
peak. Importantly, for peak variables, a value needs to be assigned to each
individual peak, even it it is `NA` (the `lengths()` of the new peak variable
must match `lengths()` of `mz` or `intensity`, i.e. the number of peaks per
spectrum).

```{r}
## Extract the full data from a spectrum
spd <- spectraData(sps, columns = union(spectraVariables(sps),
                                        peaksVariables(sps)))
## Add a new column with a *annotation* for each peak
spd$peak_anno <- list(c("a", NA_character_, "b", "c", "d"),
                      c("a", "b", "c", "d", "e", "f", "g"),
                      c("a", "b", "c", "d", "e", "f", "g", "h", "i"))
## lengths have to match:
lengths(spd$peak_anno)
lengths(spd$mz)
```

The parameter `peaksVariables()` (currently only available for the
`backendInitialize()` method of `MsBackendMemory` and `MsBackendDataFrame`)
allows to define which of the columns from an input data contain peaks variables
(in our case `"mz"`, `"intensity"` and the additional `"peak_anno"` column).

```{r}
sps2 <- Spectra(spd, backend = MsBackendMemory(),
                peaksVariables = c("mz", "intensity", "peak_anno"))
peaksVariables(sps2)
```

Full peak data can be extracted with the `peaksData()` function that has a
second parameter `columns` allowing to define which peak variables to
return. Below we extract the peak data for the second spectrum.

```{r}
peaksData(sps2, columns = peaksVariables(sps2))[[2L]]
```

We can also use the `peaksData()` function to extract the values for individual
peak variables.

```{r}
## Peak annotations for the first spectrum
peaksData(sps2, "peak_anno")[[1L]]
## Peak annotations for the second spectrum
peaksData(sps2, "peak_anno")[[2L]]
```

Peak variables can also be extracted using the `$` method:

```{r}
sps2$peak_anno
```

Similar to spectra variables it is also possible to replace values for
**existing** peaks variables using the `$<-` function.


## Filtering, aggregating and merging spectra data

Various functions are available to filter, subset and merge `Spectra`
objects. These can be generally subdivided into functions that subset or filter
*spectra data* and operations that filter *mass peak data*. A third category of
function allows to aggregate data within a `Spectra` or to merge and combine
multiple `Spectra` objects into one. Functions of the various categories are
described in the following subsections. Please refer to the function's
documentation for more details and information.


### Filter spectra data

These functions comprise subset operations that reduce the total number of
spectra in a `Spectra` object as well as filter functions that reduce the
content of the `Spectra`'s spectra data (i.e. the content of its
`spectraVariables()`). These functions thus don't change or affect the mass
peaks data of the `Spectra`'s individual spectra.

- `[`: operation to reduce a `Spectra` object to selected elements.
- `dropNaSpectraVariables()`: drops `spectraVariables()` that contain only
  missing values. The function returns a `Spectra` object with the same number
  of elements, but with eventually fewer spectra variables.
- `filterAcquisitionNum()`: retains spectra with certain acquisition numbers.
- `filterDataOrigin()`: subsets to spectra from specific origins.
- `filterDataStorage()`: subsets to spectra from certain data storage files.
- `filterEmptySpectra()`: removes spectra without mass peaks.
- `filterIsolationWindow()`: keeps spectra with the provided `mz` in their
  isolation window (m/z range).
- `filterMsLevel()`: filters by MS level.
- `filterPolarity()`: filters by polarity.
- `filterPrecursorCharge()`: retains (MSn) spectra with specified
  precursor charge(s).
- `filterPrecursorIsotopes()`: identifies precursor ions (from fragment spectra)
  that could represent isotopes of the same molecule. For each of these spectra
  groups only the spectrum of the monoisotopic precursor ion is returned. MS1
  spectra are returned without filtering.
- `filterPrecursorMaxIntensity()`: filters spectra keeping, for groups of
  spectra with similar precursor m/z, the one spectrum with the highest
  precursor intensity. All MS1 spectra are returned without filtering.
- `filterPrecursorMzRange()`: retains (MSn) spectra with a precursor m/z within
  the provided m/z range.
- `filterPrecursorMzValues(()`: retains (MSn) spectra with precursor m/z value
  matching the provided value(s) considering also a `tolerance` and `ppm`.
- `filterPrecursorScan()`: retains (parent and children) scans of an acquisition
  number.
- `filterRanges()`: filters a `Spectra` object based on (multiple) user
  defined *numeric* ranges for one or more available (numeric) spectra
  variables.
- `filterRt()`: filters based on retention time range.
- `filterValues()`: filters a `Spectra` object based on similarities of
  *numeric* values of one or more available spectra variables.
- `selectSpectraVariables()`: reduces the (spectra) data within the object to
  the selected spectra variables.


### Filter or aggregate mass peak data

These function filter or aggregate the mass peak data (`peaksData()`) of each
spectrum in a `Spectra` without changing the total number of spectra.

- `combinePeaks()`: groups peaks **within each spectrum** based on similarity of
  their m/z values and combines these into a single peak per peak group.
- `deisotopeSpectra()`: deisotopes each individual spectrum keeping only the
  monoisotopic peak for peaks groups of potential isotopologues.
- `filterFourierTransformArtefacts()`: removes (Orbitrap) fast fourier transform
  artifact peaks from spectra.
- `filterIntensity()`: filter each spectrum keeping only peaks with intensities
  meeting certain criteria.
- `filterMzRange()`: filters mass peaks keeping (or removing) those with an
  m/z within the provided m/z range.
- `filterMzValues()`: filters mass peaks within each spectrum keeping (or
  removing) those with an m/z matching the provided value(s).
- `filterPeaksRanges()`: filters mass peaks using any set of range-based filters
  on numeric spectra or peaks variables.
- `filterPrecursorPeaks()`: removes peaks with either an m/z value matching the
  precursor m/z of the respective spectrum (with parameter `mz = "=="`) or peaks
  with an m/z value larger or equal to the precursor m/z (with parameter
  `mz = ">="`).
- `reduceSpectra()`: filters individual spectra keeping only the largest peak
  for groups of peaks with similar m/z values.


### Merging, aggregating and splitting


- `c()`: combine several `Spectra` into a single `Spectra` object.
- `combineSpectra()`: allows to combine the MS data from sets of spectra into a
  single spectrum per set. Thus, instead of filtering the data, this function
  aggregates it.
- `joinSpectraData()`: merge a `DataFrame` to the existing spectra data.
- `split()`: splits the `Spectra` object based on a provided grouping factor.



### Examples and use cases for filter operations

In this example, we use the `filterValues()` function to retain spectra with a
base peak m/z close to 100 (+/- 30 ppm) and a retention time around 230 (+/- 5
s).

```{r}
sps_sub <- filterValues(sps_sciex, spectraVariables = c("basePeakMZ", "rtime"),
                        values = c(123.089, 230), tolerance = c(0,5),
                        ppm = c(30, 0), match = "all")
length(sps_sub)
```

Then, we demonstrate the usage of the `filterRanges()` function to filter
spectra based on ranges of values for variables such as base peak m/z, peak
count, and retention time.

```{r}
sps_ranges <- filterRanges(sps_sciex,
                           spectraVariables = c("basePeakMZ","peaksCount",
                                                "rtime"),
                           ranges = c(123.09,124, 3500, 3520, 259, 260),
                           match = "all")
length(sps_ranges)
```

Only one spectrum matches all the ranges. Another option for `filterValues()`
and `filterRanges()` is to use the parameter `match = "any"`, which retains
spectra that match any one of the conditions instead of having to match all of
them.  Let's run the code once again but change the match parameter this time:

```{r}
sps_ranges <- filterRanges(sps_sciex,
                           spectraVariables = c("basePeakMZ",
                                                "peaksCount", "rtime"),
                           ranges = c(123.09, 124, 3500, 3520, 259, 260),
                           match = "any")
length(sps_ranges)
```

We can see many more spectra passed the filtering step this time.

In the example below we use specific functions to select all spectra measured
in the second mzML file and subsequently filter them to retain spectra measured
between 175 and 189 seconds in the measurement run.

```{r filterfile-filterrt}
fls <- unique(dataOrigin(sps_sciex))
file_2 <- filterDataOrigin(sps_sciex, dataOrigin = fls[2])
length(file_2)

sps_sub <- filterRt(file_2, rt = c(175, 189))
length(sps_sub)
```

In addition, `Spectra` support also subsetting with `[`. Below we perform the
filtering above with `[` -based subsetting.

```{r subset-square-bracket}
sps_sciex[sps_sciex$dataOrigin == fls[2] &
          sps_sciex$rtime >= 175 &
          sps_sciex$rtime <= 189]
```

The equivalent using filter function is shown below, with the added
benefit that the filtering is recorded in the processing slot.

```{r subset-filter-pipes}
sps_sciex |>
    filterDataOrigin(fls[2]) |>
    filterRt(c(175, 189))
```

Note that the use of the filter functions might be more efficient for
some backends, depending on their implementation, (e.g. database-based
backends could *translate* the filter function into a SQL condition to
perform the subsetting already within the database).

Multiple `Spectra` objects can also be combined into a single `Spectra` with the
`c()` or the `concatenateSpectra()` function. The resulting `Spectra` object
will contain an union of the spectra variables of the individual objects. Below
we combine the `Spectra` object `sps` with an additional object containing
another MS2 spectrum for Caffeine.

```{r caf}
caf_df <- DataFrame(msLevel = 2L, name = "Caffeine",
                    id = "HMDB0001847",
                    instrument = "Agilent 1200 RRLC; Agilent 6520 QTOF",
                    splash = "splash10-0002-0900000000-413259091ba7edc46b87",
                    centroided = TRUE)
caf_df$mz <- list(c(110.0710, 138.0655, 138.1057, 138.1742, 195.9864))
caf_df$intensity <- list(c(3.837, 32.341, 0.84, 0.534, 100))

caf <- Spectra(caf_df)
```

Next we combine the two objects.

```{r combine}
sps <- concatenateSpectra(sps, caf)
sps
```

The resulting object contains now the data for all 4 MS2 spectra and an union of
all spectra variables from both objects.

```{r merge-spectravariables}
spectraVariables(sps)
```

The second object had an additional spectra variable *instrument* that was not
present in `sps` and all the spectra in this object will thus get a value of
`NA` for this variable.

```{r merge-add-column}
sps$instrument
```

Sometimes not all spectra variables might be required (e.g. also because many of
them are empty). This might be specifically interesting also for `Spectra`
containing the data from very large experiments, because it can significantly
reduce the object's size in memory. In such cases the `selectSpectraVariables()`
function can be used to retain only specified spectra variables.


## Data manipulations

Some analyses require manipulation of the mass peak data (i.e. the m/z and/or
intensity values). One example would be to remove all peaks from a spectrum that
have an intensity lower than a certain threshold. Below we perform such an
operation with the `replaceIntensitiesBelow()` function to replace peak
intensities below 10 in each spectrum in `sps` with a value of 0.

```{r replaceintensities}
sps_rep <- replaceIntensitiesBelow(sps, threshold = 10, value = 0)
```

As a result intensities below 10 were set to 0 for all peaks.

```{r replaceintensities-intensity}
intensity(sps_rep)
```

Zero-intensity peaks (and peaks with missing intensities) can then be removed
with the `filterIntensity()` function specifying a lower required intensity level
or optionally also an upper intensity limit.

```{r clean}
sps_rep <- filterIntensity(sps_rep, intensity = c(0.1, Inf))
```

```{r clean-intensity}
intensity(sps_rep)
```

The `filterIntensity()` supports also a user-provided function to be passed with
parameter `intensity` which would allow e.g. to remove peaks smaller than the
median peak intensity of a spectrum. See examples in the `?filterIntensity` help
page for details.

Note that any data manipulations on `Spectra` objects are not immediately
applied to the peak data. They are added to a so called *processing queue* which
is applied each time peak data is accessed (with the `peaksData()`, `mz()` or
`intensity()` functions). Thanks to this processing queue data manipulation
operations are also possible for *read-only* backends (e.g. mzML-file based
backends or database-based backends). The information about the number of such
processing steps can be seen below (next to *Lazy evaluation queue*).

```{r processing-queue}
sps_rep
```

It is possible to add also custom functions to the processing queue of a
`Spectra` object. Such a function must take a peaks matrix as its first
argument, have `...` in the function definition and must return a peaks matrix
(a peaks matrix is a numeric two-column matrix with the first column containing
the peaks' m/z values and the second the corresponding intensities). Below we
define a function that divides the intensities of each peak by a value which can
be passed with argument `y`.

```{r define-function}
## Define a function that takes a matrix as input, divides the second
## column by parameter y and returns it. Note that ... is required in
## the function's definition.
divide_intensities <- function(x, y, ...) {
    x[, 2] <- x[, 2] / y
    x
}

## Add the function to the procesing queue
sps_2 <- addProcessing(sps_rep, divide_intensities, y = 2)
sps_2
```

Object `sps_2` has now 3 processing steps in its lazy evaluation queue. Calling
`intensity()` on this object will now return intensities that are half of the
intensities of the original objects `sps`.

```{r custom-processing}
intensity(sps_2)
intensity(sps_rep)
```

Alternatively we could define a function that returns the maximum peak from each
spectrum (note: we use the `unname()` function to remove any names from the
results):

```{r return-max-peak}
max_peak <- function(x, ...) {
    unname(x[which.max(x[, 2]), , drop = FALSE])
}

sps_2 <- addProcessing(sps_rep, max_peak)
lengths(sps_2)
intensity(sps_2)
```

Each spectrum in `sps_2` thus contains only a single peak. The parameter
`spectraVariables` of the `addProcessing()` function allows in addition to
define spectra variables that should be passed (in addition to the peaks matrix)
to the user-provided function. This would enable for example to calculate
*neutral loss* spectra from a `Spectra` by subtracting the precursor m/z from
each m/z of a spectrum (note that there would also be a dedicated
`neutralLoss()` function to perform this operation more efficiently). Our tool
example does not have precursor m/z values defined, thus we first set them to
arbitrary values. Then we define a function `neutral_loss` that calculates the
difference between the precursor m/z and the fragment peak's m/z. In addition we
need to ensure the peaks in the resulting spectra are ordered by (the delta) m/z
values. Note that, in order to be able to access the precursor m/z of the
spectrum within our function, we have to add a parameter to the function that
has the same name as the spectrum variable we want to access (in our case
`precursorMz`).

```{r set-precursor-mz}
sps_rep$precursorMz <- c(170.5, 170.5, 195.1, 195.1)

neutral_loss <- function(x, precursorMz, ...) {
    x[, "mz"] <- precursorMz - x[, "mz"]
    x[order(x[, "mz"]), , drop = FALSE]
}
```

We have then to call `addProcessing()` with `spectraVariables = "precursorMz"`
to specify that this spectra variable is passed along to our function.

```{r neutral-loss}
sps_3 <- addProcessing(sps_rep, neutral_loss,
                       spectraVariables = "precursorMz")
mz(sps_rep)
mz(sps_3)
```

As we can see, the precursor m/z was subtracted from each m/z of the respective
spectrum. A better version of the function, that only calculates neutral loss
spectra for MS level 2 spectra would be the `neutral_loss` function below. Since
we are accessing also the spectrum's MS level we have to call `addProcessing()`
adding also the spectra variable `msLevel` to the `spectraVariables`
parameter. Note however that the `msLevel` spectra variable **is by default
renamed** to `spectrumMsLevel` prior passing it to the function. We have thus to
use a parameter called `spectrumMsLevel` in the `neutral_loss` function instead
of `msLevel`.

```{r neutral-loss2}
neutral_loss <- function(x, spectrumMsLevel, precursorMz, ...) {
    if (spectrumMsLevel == 2L) {
        x[, "mz"] <- precursorMz - x[, "mz"]
        x <- x[order(x[, "mz"]), , drop = FALSE]
    }
    x
}
sps_3 <- addProcessing(sps_rep, neutral_loss,
                       spectraVariables = c("msLevel", "precursorMz"))
mz(sps_3)
```

Using the same concept it would also be possible to provide any
spectrum-specific user-defined value to the processing function. This variable
could simply be added first as a new spectra variable to the `Spectra` object
and then this variable could be passed along to the function in the same way we
passed the precursor m/z to our function above.

Another example for spectra processing potentially helpful for spectral matching
against reference fragment spectra libraries would be a function that removes
fragment peaks with an m/z matching the precursor m/z of a spectrum. Below we
define such a function that takes the peaks matrix and the precursor m/z as
input and evaluates with the `closest()` function from the
`r Biocpkg("MsCoreUtils")` whether the spectrum contains peaks with an m/z value
matching the one of the precursor (given `tolerance` and `ppm`). The returned
peaks matrix contains all peaks except those matching the precursor m/z.

```{r}
library(MsCoreUtils)
remove_precursor <- function(x, precursorMz, tolerance = 0.1, ppm = 0, ...) {
    if (!is.na(precursorMz)) {
        keep <- is.na(closest(x[, "mz"], precursorMz, tolerance = tolerance,
                             ppm = ppm, .check = FALSE))
        x[keep, , drop = FALSE]
    } else x
}
```

We can now again add this processing step to our `Spectra` object. As a result,
peaks matching the precursor m/z (with `tolerance = 0.1` and `ppm = 0`) will be
removed.

```{r}
sps_4 <- addProcessing(sps_rep, remove_precursor,
                       spectraVariables = "precursorMz")

peaksData(sps_4) |> as.list()
```

As a reference, the original peak matrices are shown below.

```{r}
peaksData(sps_rep) |> as.list()
```

Note that we can also perform a more relaxed matching of m/z values by passing a
different value for `tolerance` to the function:

```{r}
sps_4 <- addProcessing(sps_rep, remove_precursor, tolerance = 0.6,
                       spectraVariables = "precursorMz")
peaksData(sps_4) |> as.list()
```

Since all data manipulations above did not change the original intensity or m/z
values, it is possible to *restore* the original data. This can be done with the
`reset()` function which will empty the lazy evaluation queue and call the
`reset()` method on the storage backend. Below we call `reset()` on the `sps_2`
object and hence restore the data to its original state.

```{r reset}
sps_2_rest <- reset(sps_2)

intensity(sps_2_rest)
intensity(sps)
```

Finally, for `Spectra` that use a *writeable* backend, such as the
`MsBackendMemory`, `MsBackendDataFrame` or `MsBackendHdf5Peaks`, it is possible
to apply the processing queue to the peak data and write that back to the data
storage with the `applyProcessing()` function. Below we use this to make all
data manipulations on peak data of the `sps_rep` object persistent.

```{r applyProcessing}
length(sps_rep@processingQueue)

sps_rep <- applyProcessing(sps_rep)
length(sps_rep@processingQueue)
sps_rep
```

Before `applyProcessing()` the lazy evaluation queue contained 2 processing
steps, which were then applied to the peak data and *written* to the data
storage. Note that calling `reset()` **after** `applyProcessing()` can no longer
*restore* the data.


## Visualizing `Spectra`

The `Spectra` package provides the following functions to visualize spectra
data:
- `plotSpectra()`: plot each spectrum in `Spectra` in its own panel.
- `plotSpectraOverlay()`: plot multiple spectra into the **same** plot.

Below we use `plotSpectra()` to plot the 4 spectra from the `sps` object using
their names (as provided in spectra variable `"name"`) as plot titles.

```{r plotspectra, fig.width = 8, fig.height = 8}
plotSpectra(sps, main = sps$name)
```

It is also possible to label individual peaks in each plot. Below we use the m/z
value of each peak as its label. In the example we define a function that
accesses information from each spectrum (`z`) and returns a `character` for each
peak with the text that should be used as label. Parameters `labelSrt`,
`labelPos` and `labelOffset` define the rotation of the label text and its
position relative to the x and y coordinates of the peak.

```{r plotspectra-label, fig.width = 8, fig.height = 8}
plotSpectra(sps, main = sps$name,
            labels = function(z) format(mz(z)[[1L]], digits = 4),
            labelSrt = -30, labelPos = 2, labelOffset = 0.1)
```

These plots are rather busy and for some peaks the m/z values are
overplotted. Below we define a *label function* that will only indicate the m/z
of peaks with an intensity higher than 30.

```{r plotspectra-label-int, fig.width = 8, fig.height = 8}
mzLabel <- function(z) {
    z <- peaksData(z)[[1L]]
    lbls <- format(z[, "mz"], digits = 4)
    lbls[z[, "intensity"] < 30] <- ""
    lbls
}
plotSpectra(sps, main = sps$name, labels = mzLabel,
            labelSrt = -30, labelPos = 2, labelOffset = 0.1)
```

Sometimes it might be of interest to plot multiple spectra into the **same**
plot (e.g. to directly compare peaks from multiple spectra). This can be done
with `plotSpectraOverlay()` which we use below to create an *overlay-plot* of
our 4 example spectra, using a different color for each spectrum.

```{r plotspectraoverlay, fig.width = 6, fig.height = 6}
cols <- c("#E41A1C80", "#377EB880", "#4DAF4A80", "#984EA380")
plotSpectraOverlay(sps, lwd = 2, col = cols)
legend("topleft", col = cols, legend = sps$name, pch = 15)
```

Lastly, `plotSpectraMirror()` allows to plot two spectra against each other as a
*mirror plot* which is ideal to visualize spectra comparison results. Below we
plot a spectrum of 1-Methylhistidine against one of Caffeine.

```{r plotspectramirror, fig.width = 6, fig.height = 6}
plotSpectraMirror(sps[1], sps[3])
```

The upper panel shows the spectrum from 1-Methylhistidine, the lower the one of
Caffeine. None of the peaks of the two spectra match. Below we plot the two
spectra of 1-Methylhistidine and the two of Caffeine against each other matching
peaks with a `ppm` of 50.

```{r plotspectramirror-ppm, fig.width = 12, fig.height = 6}
par(mfrow = c(1, 2))
plotSpectraMirror(sps[1], sps[2], main = "1-Methylhistidine", ppm = 50)
plotSpectraMirror(sps[3], sps[4], main = "Caffeine", ppm = 50)
```

See also `?plotSpectra` for more plotting options and examples.


## Aggregating spectra data

The `Spectra` package provides the `combineSpectra()` function that allows to
*aggregate* multiple spectra into a single one. The main parameters of this
function are `f`, which defines the sets of spectra that should be combined, and
`FUN`, which allows to define the function that performs the actual
aggregation. The default aggregation function is `combinePeaksData()` (see
`?combinePeaksData` for details) that combines multiple spectra into a single
spectrum with all peaks from all input spectra (with additional paramter `peaks
= "union"`), or peaks that are present in a certain proportion of input spectra
(with parameter `peaks = "intersect"`; parameter `minProp` allows to define the
minimum required proportion of spectra in which a peak needs to be present. It
is important to mention that, by default, the function combines all mass peaks
from all spectra with a similar m/z value into a single, representative mass
peak aggregating all their intensities into one. To avoid the resulting
intensity to be affected by potential noise peaks it might be advised to first
*clean* the individual mass spectra using e.g. the `combinePeaks()` or
`reduceSpectra()` functions that first aggregate mass peaks **within** each
individual spectrum.

In this example we below we use `combineSpectra()` to combine the spectra for
1-methylhistidine and caffeine into a single spectrum for each compound. We use
the spectra variable `$name`, that contains the names of the compounds, to
define which spectra should be grouped together.

```{r}
sps_agg <- combineSpectra(sps, f = sps$name)
```

As a result, the 4 spectra got aggregated into two.

```{r, fig.width = 4, fig.height = 8}
plotSpectra(sps_agg, main = sps_agg$name)
```

By default, all peaks present in all spectra are reported. As an alternative, by
specifying `peaks = "intersect"` and `minProp = 1`, we could combine the spectra
keeping only peaks that are present in **both** input spectra.

```{r, fig.width = 4, fig.height = 8}
sps_agg <- combineSpectra(sps, f = sps$name, peaks = "intersect", minProp = 1)
plotSpectra(sps_agg, main = sps_agg$name)
```

This results thus in a single peak for 1-methylhistidine and none for caffeine -
why? The reason for that is that the difference of the peaks' m/z values is
larger than the default tolerance used for the peak grouping (the defaults for
`combinePeaksData()` is `tolerance = 0` and `ppm = 0`). We could however already
see in the previous section that the reported peaks' m/z values have a larger
measurement error (most likely because the fragment spectra were measured on
different instruments with different precision). Thus, we next increase the
`tolerance` and `ppm` parameters to group also peaks with a larger difference in
their m/z values.

```{r, fig.width = 4, fig.height = 8}
sps_agg <- combineSpectra(sps, f = sps$name, peaks = "intersect",
                          minProp = 1, tolerance = 0.2)
plotSpectra(sps_agg, main = sps_agg$name)
```

Whether in a real analysis we would be OK with such a large tolerance is however
questionable. Note: which m/z and intensity is reported for the aggregated
spectra can be defined with the parameters `intensityFun` and `mzFun` of
`combinePeaksData()` (see `?combinePeaksData` for more information).

While the `combinePeaksData()` function is indeed helpful to combine peaks from
different spectra, the `combineSpectra()` function would in addition also allow
us to provide our own, custom, peak aggregation function. As a simple example,
instead of combining the spectra, we would like to select one of the input
spectra as *representative* spectrum for grouped input
spectra. `combineSpectra()` supports any function that takes a list of peak
matrices as input and returns a single peak matrix as output. We thus define
below a function that calculates the total signal (TIC) for each input peak
matrix, and returns the one peak matrix with the largest TIC.

```{r}
#' function to select and return the peak matrix with the largest tic from
#' the provided list of peak matrices.
maxTic <- function(x, ...) {
    tic <- vapply(x, function(z) sum(z[, "intensity"], na.rm = TRUE),
                  numeric(1))
    x[[which.max(tic)]]
}
```

We can now use this function with `combineSpectra()` to select for each compound
the spectrum with the largest TIC.

```{r, fig.width = 4, fig.height = 8}
sps_agg <- combineSpectra(sps, f = sps$name, FUN = maxTic)
plotSpectra(sps_agg, main = sps_agg$name)
```


## Comparing spectra

Spectra can be compared with the `compareSpectra()` function, that allows to
calculate similarities between spectra using a variety of
methods. `compareSpectra()` implements similarity scoring as a two-step
approach: first the peaks from the pair of spectra that should be compared are
matched (mapped) against each other and then a similarity score is calculated on
these. The `MAPFUN` parameter of `compareSpectra()` defines the function to
match (or map) the peaks between the spectra and parameter `FUN` specifies the
function to calculate the similarity. By default, `compareSpectra()` uses
`MAPFUN = joinPeaks` (see `?joinPeaks` for a description and alternative
options) and `FUN = ndotproduct` (the normalized dot-product spectra similarity
score). Parameters to configure these functions can be passed to
`compareSpectra()` as additional parameter (such as e.g. `ppm` to define the
m/z-relative tolerance for peak matching in `joinPeaks()`).

Below we calculate pairwise similarities between all spectra in `sps` accepting
a 50 ppm difference of peaks' m/z values for being considered matching.

```{r comparespectra}
compareSpectra(sps, ppm = 50)
```

The resulting matrix provides the similarity scores from the pairwise
comparison. As expected, the first two and the last two spectra are similar,
albeit only moderately, while the spectra from 1-Methylhistidine don't share any
similarity with those of Caffeine. Similarities *between* `Spectra` objects can
be calculated with calls in the form of `compareSpectra(a, b)` with `a` and `b`
being the two `Spectra` objects to compare. As a result a *n x m* matrix will be
returned with *n* (rows) being the spectra in `a` and *m* (columns) being the
spectra in `b`.

The above similarity was calculated with the default (normalized) dot-product,
but also other similarity scores can be used instead. Either one of the other
metrics provided by the `r Biocpkg( "MsCoreUtils")` could be used (see
`?MsCoreUtils::distance` for a list of available options) or any other external
or user-provided similarity scoring function. As an example, we use below the
spectral entropy similarity score introduced in [@y_spectral_2021] and provided
with the
[*msentropy*](https://cran.r-project.org/web/packages/msentropy/index.html)
package. Since this `msentropy_similarity()` function performs also the mapping
of the peaks between the compared spectra internally (along with some spectra
cleaning), we have to disable that in the `compareSpectra()` function using
`MAPFUN = joinPeaksNone`. To configure the similarity scoring we can pass all
additional parameters of the `msentropy_similarity()` (see
`?msentropy_similarity`) to the `compareSpectra()` call. We use
`ms2_tolerance_in_ppm = 50` to set the tolerance for m/z-relative peak matching
(equivalent to `ppm = 50` used above) and `ms2_tolerance_in_da = -1` to disable
absolute tolerance matching.

```{r}
library(msentropy)
compareSpectra(sps, MAPFUN = joinPeaksNone, FUN = msentropy_similarity,
               ms2_tolerance_in_ppm = 50, ms2_tolerance_in_da = -1)
```

Note also that GNPS-like scores can be calculated with `MAPFUN = joinPeaksGnps`
and `FUN = MsCoreUtils::gnps`. For additional information and examples see also
[@rainer_modular_2022] or the
[SpectraTutorials](https://jorainer.github.io/SpectraTutorials) tutorial.

Another way of comparing spectra would be to *bin* the spectra and to cluster
them based on similar intensity values. Spectra binning ensures that the binned
m/z values are comparable across all spectra. Below we bin our spectra using a
bin size of 0.1 (i.e. all peaks with an m/z smaller than 0.1 are aggregated into
one binned peak. Below, we explicitly set `zero.rm = FALSE` to retain all bins
generated by the function, including those with an intensity of zero.

```{r}
sps_bin <- Spectra::bin(sps, binSize = 0.1, zero.rm = FALSE)
```

All spectra will now have the same number of m/z values.

```{r}
lengths(sps_bin)
```

Most of the intensity values for these will however be 0 (because in the
original spectra no peak for the respective m/z bin was present).

```{r}
intensity(sps_bin)
```

We're next creating an intensity matrix for our `Spectra` object, each row being
one spectrum and columns representing the binned m/z values.

```{r}
intmat <- do.call(rbind, intensity(sps_bin))
```

We can now identify those columns (m/z bins) with only 0s across all spectra and
remove these.

```{r}
zeros <- colSums(intmat) == 0
intmat <- intmat[, !zeros]
intmat
```

The associated m/z values for the bins can be extracted with `mz()` from the
binned `Spectra` object. Below we use these as column names for the intensity
matrix.

```{r}
colnames(intmat) <- mz(sps_bin)[[1L]][!zeros]
```

This intensity matrix could now for example be used to cluster the spectra based
on their peak intensities.

```{r}
heatmap(intmat)
```

As expected, the first 2 and the last 2 spectra are more similar and are
clustered together.


## Exporting spectra

Spectra data can be exported with the `export()` method. This method takes the
`Spectra` that is supposed to be exported and the backend (parameter `backend`)
which should be used to export the data and additional parameters for the export
function of this backend. The backend thus defines the format of the exported
file. Note however that not all `MsBackend` classes might support data export.
The backend classes currently supporting data export and its format are:
- `MsBackendMzR` (`Spectra` package): export data in *mzML* and *mzXML*
  format. Can not export all custom, user specified spectra variables.
- `MsBackendMgf`
  ([`MsBackendMgf`](https://RforMassSpectrometry.github.io/MsBackendMgf)
  package): exports data in *Mascot Generic Format* (mgf). Exports all spectra
  variables as individual spectrum fields in the mgf file.
- `MsBackendMsp`
  ([`MsBackendMsp`](https://RforMassSpectrometry.github.io/MsBackendMsp)):
  exports data in NIST MSP format.
- `MsBackendMassbank`
  ([`MsBackendMassbank`](https://RforMassSpectrometry.github.io/MsBackendMassbank))
  exports data in Massbank text file format.

In the example below we use the `MsBackendMzR` to export all spectra from the
variable `sps` to an mzML file. We thus pass the data, the backend that should
be used for the export and the file name of the result file (a temporary file)
to the `export()` function (see also the help page of the `export,MsBackendMzR`
function for additional supported parameters).

```{r export}
fl <- tempfile()
export(sps, MsBackendMzR(), file = fl)
```

To evaluate which of the spectra variables were exported, we load the exported
data again and identify spectra variables in the original file which could not
be exported (because they are not defined variables in the mzML standard).

```{r export-import}
sps_im <- Spectra(backendInitialize(MsBackendMzR(), fl))
spectraVariables(sps)[!spectraVariables(sps) %in% spectraVariables(sps_im)]
```

These additional variables were thus not exported. How data export is performed
and handled depends also on the used backend. The `MsBackendMzR` for example
exports all spectra by default to a single file (specified with the `file`
parameter), but it allows also to specify for each individual spectrum in the
`Spectra` to which file it should be exported (parameter `file` has thus to be
of length equal to the number of spectra). As an example we export below the
spectrum 1 and 3 to one file and spectra 2 and 4 to another.

```{r export-twofiles}
fls <- c(tempfile(), tempfile())
export(sps, MsBackendMzR(), file = fls[c(1, 2, 1, 2)])
```

A more realistic use case for mzML export would be to export MS data after
processing, such as smoothing (using the `smooth()` function) and centroiding
(using the `pickPeaks()` function) of raw profile-mode MS data.


## Changing backends

In the previous sections we learned already that a `Spectra` object can use
different backends for the actual data handling. It is also possible to
change the backend of a `Spectra` to a different one with the `setBackend()`
function. We could for example change the (`MsBackendMzR`) backend of the
`sps_sciex` object to a `MsBackendMemory` backend to enable use of the data
even without the need to keep the original mzML files. Below we change the
backend of `sps_sciex` to the in-memory `MsBackendMemory` backend.

```{r setbackend}
print(object.size(sps_sciex), units = "Mb")
sps_sciex <- setBackend(sps_sciex, MsBackendMemory())
sps_sciex
```

With the call the full peak data was imported from the original mzML files into
the object. This has obviously an impact on the object's size, which is now much
larger than before.

```{r memory-after-import}
print(object.size(sps_sciex), units = "Mb")
```

The `dataStorage` spectrum variable has now changed, while `dataOrigin` still
keeps the information about the originating files:

```{r new-datastorage}
head(dataStorage(sps_sciex))
head(basename(dataOrigin(sps_sciex)))
```


# Backends

Backends allow to use different *backends* to store mass spectrometry data while
providing *via* the `Spectra` class a unified interface to use that data. This
is a further abstraction to the *on-disk* and *in-memory* data modes from
`MSnbase` [@gattoMSnbaseEfficientElegant2020a]. The `Spectra` package defines a
set of example backends but any object extending the base `MsBackend` class
could be used instead. The default backends are:

- `MsBackendMemory`: the *default* backend to store data in memory. Due to its
  design the `MsBackendMemory` provides fast access to the peaks matrices (using
  the `peaksData()` function) and is also optimized for fast access to spectra
  variables and subsetting. Since all data is kept in memory, this backend has a
  relatively large memory footprint (depending on the data) and is thus not
  suggested for very large MS experiments.

- `MsBackendDataFrame`: the mass spectrometry data is stored (in-memory) in a
  `DataFrame`. Keeping the data in memory guarantees high performance but has
  also, depending on the number of mass peaks in each spectrum, a much higher
  memory footprint.

- `MsBackendMzR`: this backend keeps only general spectra variables in memory
  and relies on the `r Biocpkg("mzR")` package to read mass peaks (m/z and
  intensity values) from the original MS files on-demand.

- `MsBackendHdf5Peaks`: similar to `MsBackendMzR` this backend reads peak data
  only on-demand from disk while all other spectra variables are kept in
  memory. The peak data are stored in Hdf5 files which guarantees scalability.

All of the above mentioned backends support changing all of their their spectra
variables, **except** the `MsBackendMzR` that does not support changing m/z or
intensity values for the mass peaks.

With the example below we load the data from a single mzML file and use a
`MsBackendHdf5Peaks` backend for data storage. The `hdf5path` parameter allows
us to specify the storage location of the HDF5 file.

```{r hdf5}
library(msdata)
fl <- proteomics(full.names = TRUE)[5]

sps_tmt <- Spectra(fl, backend = MsBackendHdf5Peaks(), hdf5path = tempdir())
head(basename(dataStorage(sps_tmt)))
```

A (possibly incomplete) list of R packages providing additional backends that
add support for additional data types or storage options is provided below:

- `MsBackendCompDb` (package `r BiocStyle::Biocpkg("CompoundDb")`): provides
  access to spectra data (spectra and peaks variables) from a *CompDb*
  database. Has a small memory footprint because all data (except precursor m/z
  values) are retrieved on-the-fly from the database.

- `MsBackendHmdbXml` (package
  [`MsbackendHmdb`](https://github.com/rformassspectrometry/MsBackendHmdb)):
  allows import of MS data from xml files of the Human Metabolome Database
  (HMDB). Extends the `MsBackendDataFrame` and keeps thus all data, after
  import, in memory.

- `MsBackendMassbank` (package `r BiocStyle::Biocpkg("MsBackendMassbank")`):
  allows to import/export data in MassBank text file format. Extends the
  `MsBackendDataFrame` and keeps thus all data, after import, in memory.

- `MsBackendMassbankSql` (package `r BiocStyle::Biocpkg("MsBackendMassbank")`):
  allows to directly connect to a MassBank SQL database to retrieve all MS data
  and variables. Has a minimal memory footprint because all data is retrieved
  on-the-fly from the SQL database.

- `MsBackendMetaboLights` (package `r
  BiocStyle::Biocpkg("MsBackendMetaboLights")`): retrieves and caches MS data
  files from the MetaboLights repository.

- `MsBackendMgf`: (package `r BiocStyle::Biocpkg("MsBackendMgf")`): support for
  import/export of mass spectrometry files in mascot generic format (MGF).

- `MsBackendMsp`: (package `r BiocStyle::Biocpkg("MsBackendMsp")`): allows to
  import/export data in NIST MSP format. Extends the `MsBackendDataFrame` and
  keeps thus all data, after import, in memory.

- `MsBackendRawFileReader` (package `r Biocpkg("MsBackendRawFileReader")`):
  implements a backend for reading MS data from Thermo Fisher Scientific's raw
  data files using the manufacturer's NewRawFileReader .Net libraries. The
  package generalizes the functionality introduced by the `r Biocpkg("rawrr")`
  package, see also [@kockmann_rawrr_2021].

- `MsBackendSql` (package `r BiocStyle::Biocpkg("MsBackendSql")`): stores all MS
  data in a SQL database and has thus a minimal memory footprint.

- `MsBackendTimsTof` (package
  [`MsBackendTimsTof`](https://github.com/rformassspectrometry/MsBackendTimsTof):
  allows import of data from Bruker TimsTOF raw data files (using the
  `opentimsr` R package).

- `MsBackendWeizMass` (package
  [`MsBackendWeizMass`](https://github.com/rformassspectrometry/MsBackendWeizMass):
  allows to access MS data from WeizMass MS/MS spectral databases.


# Handling very large data sets

The `Spectra` package was designed to support also efficient processing of very
large data sets. Most of the functionality do not require to keep the full MS
data in memory (specifically, the peaks data, i.e., m/z and intensity values,
which represent the largest chunk of data for MS experiments). For some
functions however the peaks data needs to be loaded into memory. One such
example is the `lengths()` function to determine the number of peaks per spectra
that is calculated (on the fly) by evaluating the number of rows of the peaks
matrix. Backends such as the `MsBackendMzR` perform by default any data
processing separately (and eventually in parallel) by data file and it should
thus be safe to call any such functions on a `Spectra` object with that
backend. For other backends (such as the
[`MsBackendSql`](https://github.com/RforMassSpectrometry/MsBackendSql) or the
[`MsBackendMassbankSql`](https://github.com/RforMassSpectrometry/MsBackendMassbank))
it is however advised to process the data in a *chunk-wise* manner using the
`spectrapply()` function with parameter `chunkSize`. This will split the
original `Spectra` object into chunks of size `chunkSize` and applies the
function separately to each chunk. That way only data from one chunk will
eventually be loaded into memory in each iteration enabling to process also very
large `Spectra` objects on computers with limited hardware resources. Instead of
a `lengths(sps)` call, the number of peaks per spectra could also be determined
(in a less memory demanding way) with `spectrapply(sps, lengths, chunkSize =
5000L)`. In that way only peak data of 5000 spectra at a time will be loaded
into memory.


# Serializing (saving), moving and loading serialized `Spectra` objects

Serializing and re-loading variables/objects during an analysis using e.g. the
`save()` and `load()` functions are common in many workflows, especially if some
of the tasks are computationally intensive and take long time. Sometimes such
serialized objects might even be moved from one computer (or file system) to
another. These operations are unproblematic for `Spectra` objects with
*in-memory* backends such as the `MsBackendMemory` or `MsBackendDataFrame`, that
keep all data in memory, would however break for *on-disk* backends such as the
`MsBackendMzR` if the file path to the original data files is not identical. It
is thus suggested (if the size of the MS data respectively the available system
memory allows it) to change the backend for such `Spectra` objects to a
`MsBackendMemory` before serializing the object with `save()`. For `Spectra`
objects with a `MsBackendMzR` an alternative option would be to eventually
update/adapt the path to the directory containing the raw (e.g. mzML) data
files: assuming these data files are available on both computers, the path to
the directory containing these can be updated with the `dataStorageBasePath<-`
function allowing thus to move/copy serialized `Spectra` objects between
computers or file systems.

An example workflow could be:

files *a.mzML*, *b.mzML* are stored in a directory */data/mzML/* on one
computer. These get loaded as a `Spectra` object with `MsBackendMzR` and then
serialized to a file *A.RData*.

```{r, eval = FALSE}
A <- Spectra(c("/data/mzML/a.mzML", "/data/mzML/b.mzML"))
save(A, file = "A.RData")
```

Assuming this file gets now copied to another computer (where the data is not
available in a folder */data/mzML/*) and loaded with `load()`.

```{r, eval = FALSE}
load("A.RData")
```

This `Spectra` object would not be valid because its `MsBackendMzR` can no
longer access the MS data in the original data files. Assuming the user also
copied the data files *a.mzML* and *b.mzML*, but to a folder
*/some_other_folder/*, the base storage path of the object would need to be
adapted to match the directory where the data files are available on the second
computer:

```{r, eval = FALSE}
dataStorageBasePath(A) <- "/some_other_folder"
```

By pointing now the storage path to the new storage location of the data files,
the `Spectra` object `A` would also be usable on the second computer.


# Session information

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

# References