toxpiR Introduction

Introduction

This document introduces ToxPi and describes how to use the toxpiR package to easily import, recombine, analyze, and visualize high dimensional data. The toxpiR package is an R implementation of ToxPi that offers new features over what was previously available for data handling, recombination, and customization; provides formally packaged, open-source code for ToxPi; extends the application domain by supporting rapid analysis of massive datasets; and bridges with the stand-alone, Graphical User Interface (GUI) Java application and ArcGIS Toolkit.

What is ToxPi?

Toxicological Priority Index (ToxPi) is a decision support tool that allows transparent integration and visualization of data across disparate information domains to aid in prioritization. ToxPi takes input data of disparate sources,from a biological assay or a computer predicted model,to genetic features or proteomic data, and combines all of these data types into one overall model. This model then calculates an overall score for each datapoint of interest. It does this by the user specifying one or more features to go into each “slice” of a unit circle, and the weights that these slices have. These slices can contain one or more features of any type in the same slice. The slice weights are user defined, and decided based on prior information that one may have about the analysis. When a slice has a higher weight, it takes up more room on the unit square. When a slice has a higher calculated score, it goes further out from the center of the circle. As shown below, a feature with a low score will have overall smaller slices than one with an overall higher score. This means that we can understand at a glance the top level differences between what we are interested in. The component slices also add up to the overall ToxPi score with the weights add importance to that specific slice. The metrics that make up each component slice are simulated in data in this case, but can be many different kinds of data that fits your own analysis. More information on the methodological details can be found at toxpi.org.

Outline

There are three main steps in using this package:

• Loading data: Users can load comma-separated value (csv) formatted data where data features that will make up the slices are stored as columns and attributes are rows. If you are more famililar with the GUI data format, the package also has a function to load data that is formatted in any one of the GUI layouts where slice information is presented in rows beginning with a hash tag(#). However, new users should use a simple .csv file with the row and column layout.

• Creating ToxPi Model: ToxPi model can be created by specifying information about each slice and its corresponding weights. Each slice can be made up of one or more features from the input data. You should also name the slices and provide their weights in this step.

• Computing ToxPi Scores: Scores can be computed per attribute as well as for each slice. The S4 object resulting from this step will combine and store the ToxPi model and scores all in one place. This final object will contain all the information needed for plotting.

Data contained within different slots of the S4 objects created using this package can be obtained using accessor functions. More details on how this can be done is present later in the document.

Installation

install.packages("toxpiR")

remotes::install_github("ToxPi/toxpiR")

Launching package in R

library(toxpiR)

Flowchart

This flowchart details the basic steps necessary to analyze data using toxpiR. A separate vignette describes how load and work with the GUI data file using txpImportGui() function. Usage for all other functions is described below using an example data set. To view a list of all functions available in this package, please run:

lsf.str(“package:toxpiR”)

Example usage

Data can either be entered as a regular csv file without the header lines specifying slice information, or in one of the GUI input formats. If you are using the GUI format, please refer to the Import ToxPi GUI files vignette.

The standard data input is a standard R dataframe with rows as your datapoints and columns as features that may be turned into ToxPi slices. A simple example data set called txp_example_input is available within the package. We will use this 10 x 8 toy data to walk through a simple analysis and explain the functions available in this package. These metrics that we have specified can be in completely different scales and data type, but first we will show a simple example of the package usage.

data(txp_example_input, package = "toxpiR")
head(txp_example_input)
#>     name metric1 metric2 metric3 metric4 metric5 metric6 metric7 metric8
#> 1 chem01      74      77      25      74      77      97      25      77
#> 2 chem02      28      20      72      28      20      68      72      20
#> 3 chem03      61       3      73      61       3      24      73       3
#> 4 chem04      NA      40      20      NA      40      22      20      40
#> 5 chem05      29      53      44      29      53       4      44      53
#> 6 chem06      12      43      83      12      43      85      83      43

## Goal - Create two slices with transformation functions 
# Slice 1: Metric 1, No transformation 
# Slice 2: Metric 2 (square) and Metric 3 (no transformation)

slice2.trans <- TxpTransFuncList(func1 = function(x) x^2, func2 = NULL)

f.slices <- TxpSliceList(Slice1 = TxpSlice("metric1"), 
                         Slice2 = TxpSlice(c("metric2", "metric3"), 
                                           txpTransFuncs = slice2.trans ))

## Goal - Create ToxPi model.
# Slice 1, weight = 2
# Slice 2, weight = 1, apply log transform to final value. 

# Object storing list of transformation functions.
final.trans <- TxpTransFuncList(f1 = NULL, f2 = function(x) log10(x)) 

f.model <- TxpModel(txpSlices = f.slices, 
                    txpWeights = c(2,1),
                    txpTransFuncs = final.trans)

f.results <- txpCalculateScores(model = f.model, 
                                input = txp_example_input,
                                id.var = 'name' ) 

txpSliceScores(f.results) #ToxPi scores
#>          Slice1    Slice2
#>  [1,] 0.6666667 0.3200390
#>  [2,] 0.1720430 0.1307208
#>  [3,] 0.5268817 0.0000000
#>  [4,] 0.0000000 0.2228243
#>  [5,] 0.1827957 0.2650928
#>  [6,] 0.0000000 0.2359788
#>  [7,] 0.1827957 0.0000000
#>  [8,] 0.4946237 0.2978895
#>  [9,] 0.4193548 0.3333333
#> [10,] 0.5161290 0.2233755
txpWeights(f.results) #Print weights
#> [1] 2 1

library(grid) # Load library
plot(f.results) # ToxPi visuals

# grid.ls() #List grid info

# Highlight one figure using its label
grid.edit("pie-1", fills = c("red", "black"))

# Or just one slice in a figure
grid.edit("pie-10::slice1", gp = gpar(fill = "#7DBC3D"))

# Rank plot
plot(f.results, y = txpRanks(f.results), labels = 1:10)

# Hierarchical Clustering

f.hc <- hclust(dist(txpSliceScores(f.results)))

plot(f.hc, hang = -1, labels = txpIDs(f.results), xlab = '', sub = '')

Citation

To cite toxpiR in your work, please use: https://cran.r-project.org/package=toxpiR (Manuscript currently in review)