---
title: "XCMS Parameter Optimization with IPO"
author: "Gunnar Libiseller, Thomas Riebenbauer<br>JOANNEUM RESEARCH Forschungsgesellschaft m.b.H., Graz, Austria"
date: "`r Sys.Date()`"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{XCMS Parameter Optimization with IPO}
  %\VignetteEngine{knitr::rmarkdown}
  \usepackage[utf8]{inputenc}
---

```{r, echo = FALSE, message = FALSE, warning = FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
library(faahKO)
```

## Introduction

This document describes how to use the R-package `IPO` 
to optimize `xcms` parameters. Code examples on how to
use `IPO` are provided. Additional to `IPO` the R-packages
`xcms` and `rsm` are required. The R-package `msdata` and`mtbls2` 
are recommended. The optimization process looks as following:

<br>
<b>IPO optimization process</b>
<p>
<img src="FlowChart.jpg" width="695" height="745">
</p>

## Installation

```{r install_IPO, eval=FALSE}
# try http:// if https:// URLs are not supported
source("https://bioconductor.org/biocLite.R")
biocLite("IPO")
```

<b>Installing main suggested packages</b>

```{r install_IPO_suggestions, eval=FALSE}
# for examples of peak picking parameter optimization:
biocLite("msdata")
# for examples of optimization of retention time correction and grouping
# parameters:
biocLite("faahKO")
```


## Raw data

`xcms` handles the file processing hence all files can be used 
that can be processed by `xcms`.


```{r file_choosing}
datapath <- system.file("cdf", package = "faahKO")
datafiles <- list.files(datapath, recursive = TRUE, full.names = TRUE)
```



## Optimize peak picking parameters

To optimize parameters different values (levels) have to 
tested for these parameters. To efficiently test many 
different levels design of experiment (DoE) is used. 
Box-Behnken and central composite designs set three
evenly spaced levels for each parameter. The method 
`getDefaultXcmsSetStartingParams` provides default values
for the lower and upper levels defining a range. Since
the levels are evenly spaced the middle level or center 
point is calculated automatically. To edit the starting levels
of a parameter set the lower and upper level as desired.
If a parameter should not be optimized, set a single 
default value for `xcms` processing, do not set this
parameter to NULL. 

The method `getDefaultXcmsSetStartingParams` creates a 
list with default values for the optimization of the
peak picking methods `centWave` or `matchedFilter`. To
choose between these two method set the parameter accordingly.

The method `optimizeXcmsSet` has the following parameters:

- files: the raw data which is the basis for optimization. 
  This does not necessarly need to be the whole dataset, 
  only quality controls should suffice.
- params: a list consisting of items named according to 
  `xcms` peak picking methods parameters. A default list 
  is created by `getDefaultXcmsSetStartingParams()`.
- BPPARAM: a `BiocParallelParam`-object (see `?BiocParallel::BiocParallelParam`)
  to controll the use of parallelisation of `xcms`.
  Defaults to `bpparam()`.
- nSlaves: the number of experiments of an DoE processed in parallel
- subdir: a directory where the response surface models are
  stored. Can also be `NULL` if no rsm's should be saved.

The optimization process starts at the specified levels. After
the calculation of the DoE is finished the result is 
evaluated and the levels automatically set accordingly.
Then a new DoE is generated and processed. This continues 
until an optimum is found.

The result of peak picking optimization is a list consisting
of all calculated DoEs including the used levels, design, 
response, rsm and best setting. Additionally the last list 
item is a list (`\$best_settings`) providing the optimized 
parameters (`\$parameters`), an xcmsSet object (`\$xset`) 
calculated with these parameters and the response this 
`xcms`-object gives.

```{r load_IPO, message=FALSE}
library(IPO)
```

```{r optimize_peak_picking, fig.height=7, fig.width=7, warning=FALSE}
peakpickingParameters <- getDefaultXcmsSetStartingParams('matchedFilter')
#setting levels for step to 0.2 and 0.3 (hence 0.25 is the center point)
peakpickingParameters$step <- c(0.2, 0.3)
peakpickingParameters$fwhm <- c(40, 50)
#setting only one value for steps therefore this parameter is not optimized
peakpickingParameters$steps <- 2

time.xcmsSet <- system.time({ # measuring time
resultPeakpicking <- 
  optimizeXcmsSet(files = datafiles[1:2], 
                  params = peakpickingParameters, 
                  nSlaves = 1, 
                  subdir = NULL,
                  plot = TRUE)
})
```

```{r optimize_peak_picking_result}
resultPeakpicking$best_settings$result
optimizedXcmsSetObject <- resultPeakpicking$best_settings$xset
```


The response surface models of all optimization steps for the
parameter optimization of peak picking are shown above.

<!-- <br> -->
<!-- <b>Response surface models of DoE 1 of peak picking parameter optimization</b> -->
<!-- <p><img src="rsmDirectory/rsm_1.jpg" width="695" height="521"></p> -->
<!-- <br> -->
<!-- <b>Response surface models of DoE 2 of peak picking parameter optimization</b> -->
<!-- <p><img src="rsmDirectory/rsm_2.jpg" width="695" height="521"></p> -->

Currently the `xcms` peak picking methods `centWave` 
and `matchedFilter` are supported. The parameter `peakwidth` of
the peak picking method `centWave` needs two values defining
a minimum and maximum peakwidth. These two values need separate
optimization and are therefore split into `min_peakwidth` and
`max_peakwidth` in `getDefaultXcmsSetStartingParams`. Also for
the `centWave` parameter prefilter two values have to be set.
To optimize these use set `prefilter` to optimize the first value
and `prefilter_value` to optimize the second value respectively.



## Optimize retention time correction and grouping parameters

Optimization of retention time correction and grouping
parameters is done simultaneously. The method 
`getDefaultRetGroupStartingParams` provides default 
optimization levels for the `xcms` retention time correction
method `obiwarp` and the grouping method `density`. 
Modifying these levels should be done the same way done 
for the peak picking parameter optimization.

The method `getDefaultRetGroupStartingParams` only supports
one retention time correction method (`obiwarp`) and one grouping 
method (`density`) at the moment. 

The method `optimizeRetGroup` provides the following parameter:
- xset: an `xcmsSet`-object used as basis for retention time 
  correction and grouping.
- params:  a list consisting of items named according to `xcms` 
  retention time correction and grouping methods parameters. 
  A default list is created by `getDefaultRetGroupStartingParams`.
- nSlaves: the number of experiments of an DoE processed in parallel
- subdir: a directory where the response surface models are
  stored. Can also be NULL if no rsm's should be saved.

A list is returned similar to the one returned from peak 
picking optimization. The last list item consists of the
optimized retention time correction and grouping parameters
(`\$best_settings`).


```{r optimize_retcor_group, fig.height=7, fig.width=7, warning = FALSE}
retcorGroupParameters <- getDefaultRetGroupStartingParams()
retcorGroupParameters$profStep <- 1
retcorGroupParameters$gapExtend <- 2.7
time.RetGroup <- system.time({ # measuring time
resultRetcorGroup <-
  optimizeRetGroup(xset = optimizedXcmsSetObject, 
                   params = retcorGroupParameters, 
                   nSlaves = 1, 
                   subdir = NULL,
                   plot = TRUE)
})
```


The response surface models of all optimization steps for the
retention time correction and grouping parameters are shown above.

<!-- <br> -->
<!-- <b> -->
<!--   Response surface models of DoE 1 of retention time correction and grouping  -->
<!--   parameter optimization</b> -->
<!-- <p><img src="rsmDirectory/retgroup_rsm_1.jpg" width="695" height="348"></p> -->
<!-- <br> -->
<!-- <b> -->
<!--   Response surface models of DoE 2 of retention time correction and grouping  -->
<!--   parameter optimizationn -->
<!-- </b> -->
<!-- <p><img src="rsmDirectory/retgroup_rsm_2.jpg" width="695" height="348"></p> -->
<!-- <br> -->
<!-- <b> -->
<!--   Response surface models of DoE 3 of retention time correction and grouping  -->
<!--   parameter optimization -->
<!-- </b> -->
<!-- <p><img src="rsmDirectory/retgroup_rsm_3.jpg" width="695" height="348"></p> -->

Currently the `xcms` retention time correction method
`obiwarp` and grouping method `density` are supported.



## Display optimized settings

A script which you can use to process your raw data
can be generated by using the function `writeRScript`.


```{r display_settings}
writeRScript(resultPeakpicking$best_settings$parameters, 
             resultRetcorGroup$best_settings)
```



## Running times and session info

Above calculations proceeded with following running times.
```{r times}
time.xcmsSet # time for optimizing peak picking parameters
time.RetGroup # time for optimizing retention time correction and grouping parameters

sessionInfo()
```