---
title: "Working with `DataCombined` class"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Working with `DataCombined` class}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
# Evaluate the runtime chunks only when the OSPSuite .NET runtime initialised
# successfully (native libraries + .NET). On machines without a working runtime
# the code is shown but not executed, so the vignette still renders.
.ospRuntimeAvailable <- isTRUE(tryCatch(
  requireNamespace("ospsuite", quietly = TRUE) &&
    ospsuite::getOSPSuiteSetting("initialized"),
  error = function(e) FALSE
))
knitr::opts_chunk$set(
  eval = .ospRuntimeAvailable,
  collapse = TRUE,
  warning = FALSE,
  message = FALSE,
  comment = "#>",
  out.width = "100%",
  fig.height = 6,
  fig.width = 8,
  fig.showtext = TRUE
)
```

## Introduction

In Modeling and Simulation (M&S) workflows, we often need to work with multiple 
observed and/or simulated datasets. Some of these datasets can be linked 
together in a group (e.g., the simulated and the corresponding observed data)
for analysis and visualization.

The `DataCombined` class in `{ospsuite}` provides a container to store 
data, to group them, and to transform them (e.g. scale or offset). The 
class accepts data coming from `SimulationResults` or as a `DataSet` object 
(see [Observed data](observed-data.html)).

Let's first generate some simulation results and load observed data.

```{r}
library(ospsuite)

# Simulation results
simFilePath <- system.file("extdata", "Aciclovir.pkml", package = "ospsuite")
sim <- loadSimulation(simFilePath)
simResults <- runSimulations(sim)[[1]]
outputPath <- "Organism|PeripheralVenousBlood|Aciclovir|Plasma (Peripheral Venous Blood)"

# observed data
obsData <- lapply(
  c("ObsDataAciclovir_1.pkml", "ObsDataAciclovir_2.pkml", "ObsDataAciclovir_3.pkml"),
  function(x) loadDataSetFromPKML(system.file("extdata", x, package = "ospsuite"))
)
# Name the elements of the list according to the names of the loaded `DataSet`.
names(obsData) <- lapply(obsData, function(x) x$name)
```

Typically, `DataCombined` is used to create figures using the `plotXXX()` functions. 
The different plot types are described in [Visualizations with `DataCombined`](data-combined-plotting.html). 
To visualize the different functionalities of `DataCombined` in this document, 
we will use the `plotIndividualTimeProfile()` function.

## Creating `DataCombined` object

First, we create a new instance of `DataCombined` class.

```{r}
myDataCombined <- DataCombined$new()
```

Then we add simulated results to this object using `$addSimulationResults()`:

```{r}
myDataCombined$addSimulationResults(
  simulationResults = simResults,
  quantitiesOrPaths = outputPath,
  names = "Aciclovir Plasma",
  groups = "Aciclovir PVB"
)
```

Next we add observed data to this object using `$addDataSets()`:

```{r}
myDataCombined$addDataSets(
  obsData$`Vergin 1995.Iv`,
  groups = "Aciclovir PVB"
)
```

The package exports `dataCombinedAciclovir` — a `DataCombined` object equivalent to the example built above — which can be used directly to illustrate and test how `DataCombined` objects work:

```{r}
dataCombinedAciclovir
```

Every data, be it simulated results or from `DataSet`, must have a unique name 
within `DataCombined`. If not specified by user, the path of simulated results 
or the `$name` property of the `DataSet` are used as the name. Alternatively, we 
can define the name when adding the data, as in the above example adding simulated 
results.

There are a few things to keep in mind here:

- It doesn't matter in which order observed or simulated datasets are added.
- You can add multiple `DataSet` objects in `$addDataSets()` method call.
- You can add only a single instance of `SimulationResults` in `$addSimulationResults()` method call.
- If you add a dataset with the same name as existing (e.g., adding simulation results 
from multiple simulations and not specifying the name, so the path of the results 
is used as the name), the new data will replace the existing.

## Grouping

Since the `DataCombined` object can store many datasets, some of them may naturally form a grouping and you would wish to track this in the object. Both `$addDataSets()` and `$addSimulationResults()` allow group specification via `groups` argument.

When being plotted, data sets without a grouping will appear with distinct color, 
line style (for simulated results), or symbol style (for observed data) with a 
separate entry in the legend, with the name of the data set being the legend entry.

Though it is possible to group data with different dimensions, it makes sense to 
only group data sets that have the same dimension (or dimensions that can be 
transformed into each other, like `Amount` and `Mass`) in order to be able to compare the 
data. Grouping data sets with different dimensions most likely will result in an 
error when trying to create plots or calculating residuals with such a `DataCombined`.

Let's create a `DataCombined` and add simulation results and observed data *without* 
specifying their grouping and create a time profile:

```{r}
myDataCombined <- DataCombined$new()

myDataCombined$addSimulationResults(
  simulationResults = simResults,
  quantitiesOrPaths = outputPath,
  names = "Aciclovir Plasma"
)

myDataCombined$addDataSets(
  obsData$`Vergin 1995.Iv`
)

plotIndividualTimeProfile(dataCombined = myDataCombined)
```

If you do not specify `groups` when you add datasets, and wish to update 
groupings later, you can use the `$setGroups()` method. All data within one 
group will get one entry in the legend with the name of the group, and will be 
plotted with the same color.

```{r}
myDataCombined$setGroups(
  names = c("Aciclovir Plasma", obsData$`Vergin 1995.Iv`$name),
  groups = c("Aciclovir PVB", "Aciclovir PVB")
)

plotIndividualTimeProfile(dataCombined = myDataCombined)
```

At any point, you can check the current names and groupings with the following active field:

```{r}
myDataCombined$groupMap
```

## Transformations

Sometimes the raw data included in `DataCombined` needs to be transformed using specified
offset and scale factor values.

This is supported via `$setDataTransformations()` method, where you can specify the names of datasets, and offsets and scale factors. If these arguments are scalar (i.e., of length 1), then the same value will be applied to all specified names. Scale factors are unitless values by which raw data values will 
be multiplied, offsets are added to the raw values without any conversion and must 
therefore be in the same units as the raw data.

The internal data frame in `DataCombined` will be transformed with the specified parameters, with the new columns computed as: 

  - For x values: `newXValue = (rawXValue + xOffset) * xScaleFactor`
  - For y values: `newYValue = (rawYValue + yOffset) * yScaleFactor`
  - For arithmetic error: `newErrorValue = rawErrorValue * abs(yScaleFactor)`
  - For geometric error: no transformation.

At any point, you can check the applied offsets and scale factors with the following active field:

```{r}
myDataCombined$dataTransformations
```

Now lets take a closer look on possible situations where you would apply any 
transformations. Consider the observed data from the examples above. It reports 
concentrations of aciclovir after single intravenous administration.

```{r}
myDataCombinedTranformations <- DataCombined$new()

myDataCombinedTranformations$addDataSets(
  obsData$`Vergin 1995.Iv`
)

plotIndividualTimeProfile(dataCombined = myDataCombinedTranformations)
```

However, we might want to use this data set with a simulation where aciclovir is 
administered 24 hours after simulation begin. To be able to compare simulation 
results with the data set, we can *offset*  the observed data time by 24 hours. 
Keep in mind that the offset must be given in the same unit as the data set values are.

```{r}
# Check the units of the observed time values
obsData$`Vergin 1995.Iv`$xUnit

myDataCombinedTranformations$setDataTransformations(
  forNames = obsData$`Vergin 1995.Iv`$name,
  xOffsets = 24
)

plotIndividualTimeProfile(dataCombined = myDataCombinedTranformations)
```

In the next step, we want to *normalize* observed concentrations to a dose. We 
can easily achieve this with the scale factor. In the next example, we normalize 
observed values to 250 mg dose by setting the `yScaleFactor` to 1/250:

```{r}
myDataCombinedTranformations$setDataTransformations(
  forNames = obsData$`Vergin 1995.Iv`$name,
  yScaleFactors = 1 / 250
)

plotIndividualTimeProfile(dataCombined = myDataCombinedTranformations)
```
Finally, offsetting the observation values might be useful when working with measurements 
of endogenous substrates, such as the hormone glucagon, and want to correct for the 
individual specific baseline levels of the hormone.

## Extracting a combined data frame

The data frame (also sometimes called as a table) data structure is central to R-based workflows, and, thus, we may wish to extract a data frame for datasets present in the object.

Internally, `DataCombined` extracts data frames for observed and simulated datasets and combines them.

```{r}
myDataCombined$toDataFrame()
```

This function returns a [tibble data frame](https://r4ds.had.co.nz/tibbles.html#tibbles-vs.-data.frame). If you wish to modify how it is printed, you can have a look at the available options [here](https://pillar.r-lib.org/reference/pillar_options.html). In fact, let's change a few options and print the data frame again.

```{r, eval=FALSE}
options(
  pillar.width = Inf, # show all columns
  pillar.min_chars = Inf # to turn off truncation of column titles
)

myDataCombined$toDataFrame()
```


```{r, echo=FALSE}
# change these settings only temporarily
withr::with_options(
  list(
    pillar.width = Inf, # show all columns
    pillar.min_chars = Inf # to turn off truncation of column titles
  ),
  code = {
    print(myDataCombined$toDataFrame())
  }
)
```

`{ospsuite}` also provides a few helper functions to modify the data frame further.

When multiple (observed and/or simulated) datasets are present in a data frame, they are likely to have different units. `convertUnits()` function helps to convert them to a common unit. The function 
will not modify the `DataCombined` object but return a new data frame with unified 
units.^[Note that if you are using `DataCombined` object for plotting functions, you don't need to this conversion; the plotting functions will take care of this internally.].

```{r}
convertUnits(
  myDataCombined,
  xUnit = ospUnits$Time$s,
  yUnit = ospUnits$`Concentration [mass]`$`µg/l`
)
```

## Further functionalities

Grouping of simulated results with observed data inside `DataCombined` also allows 
to calculate the error between simulated results and observations using the 
`calculateResiduals()` function. The function calculates the distance between 
each observed data point within the group and the simulated results. Simulation 
results interpolated if no simulated value is available for a certain observed 
time point. Observed data beyond simulated time is ignored, i.e., no *extrapolation* 
is performed.

There are three ways to calculate the distance (residuals) - linear, logarithmic,
and ratio - specified by the `scaling` argument. For linear scaling (`scaling = "linear"`):

> Residuals are calculated as: Simulation value - Observed value. This means that the residuals are defined by absolute differences.

For logarithmic scaling (`scaling = "log"`):

> Residuals are calculated as: log(Simulation value) - log(Observed value) = log(Simulation Value / Observed Value). Data points where the observed or predicted value is zero or negative cannot be log-transformed; these points are excluded from the output and a warning is issued.

For ratio scaling (`scaling = "ratio"`):

> Residuals are calculated as: Observed value / Simulation value.


```{r}
# Linear residuals
calculateResiduals(myDataCombined, scaling = "linear")$residualValues

# Logarithmic residuals
calculateResiduals(myDataCombined, scaling = "log")$residualValues


```

To quickly calculate the total error of the `DataCombined`, one can sum up the 
absolute values of the residuals:

```{r}
# Linear residuals
totalError <- sum(abs(calculateResiduals(myDataCombined, scaling = tlf::Scaling$lin)$residualValues))

print(totalError)
```

## Visualizations with `DataCombined`

See [Visualizations with `DataCombined`](data-combined-plotting.html) describing functions that can visualize data stored in `DataCombined` object.