Title:	Functional Data Analysis for Multiplexed Cell Images
Version:	1.1.1
Description:	Compare variables of interest between (potentially large numbers of) spatial interactions and meta-variables. Spatial variables are summarized using K, or other, functions, and projected for use in a modified random forest model. The model allows comparison of functional and non-functional variables to each other and to noise, giving statistical significance to the results. Included are preparation, modeling, and interpreting tools along with example datasets, as described in VanderDoes et al., (2023) <doi:10.1101/2023.07.18.549619>.
License:	GPL (≥ 3)
URL:	https://github.com/jrvanderdoes/funkycells, https://jrvanderdoes.github.io/funkycells/
BugReports:	https://github.com/jrvanderdoes/funkycells/issues
Depends:	R (≥ 2.10)
Imports:	fda, ggplot2, rpart, spatstat.explore, spatstat.geom, stats, stringr, tidyr
Suggests:	knitr, pROC, rmarkdown, scales, testthat (≥ 3.0.0)
VignetteBuilder:	knitr
Config/testthat/edition:	3
Encoding:	UTF-8
LazyData:	true
RoxygenNote:	7.2.3
NeedsCompilation:	no
Packaged:	2023-08-08 12:48:11 UTC; jerem
Author:	Jeremy VanderDoes [aut, cre, cph], Jack Hywood [aut], Gregory Rice [aut]
Maintainer:	Jeremy VanderDoes <jeremy.vanderdoes@gmail.com>
Repository:	CRAN
Date/Publication:	2023-08-09 16:00:06 UTC

Triple Negative Breast Cancer Data

Description

A funky model ready set of principle components from K functions based on triple negative breast cancer data from patients. The original data was proteins as coded in T/F values. Additionally, the age meta-variable was added.

Usage

TNBC

Format

TNBC

A data frame with 33 rows and 1398 columns:

Class

Outcome of each patient

Person

Person for each image

NA_Si_PC1 through tumerYN_tumerYN_PC3

Principle components of the K functions for the named interactions

age

Meta-variable for patient age

...

Source

https://www.angelolab.com/mibi-data

Triple Negative Breast Cancer Phenotypes

Description

Data of meta-variable age related to triple negative breast cancer biopsies from patients.

Usage

TNBC_meta

Format

TNBC_meta

A data frame with 33 rows and 2 columns:

Person

Person for each image

Age

Meta-variable for patient age

...

Source

https://www.angelolab.com/mibi-data

Triple Negative Breast Cancer Phenotypes

Description

Data of triple negative breast cancer biopsies from patients.

Usage

TNBC_pheno

Format

TNBC_pheno

A data frame with 170,171 rows and 5 columns:

Class

Outcome of each patient

Person

Person for which each cell is related

cellx, celly

The x-y coordinates of the cell

Phenotype

The classified phenotype for the cecll

...

Source

https://www.angelolab.com/mibi-data

Compute Pseudo-ROC Curves

Description

An receiver operating characteristic (ROC) curve is a curve showing the performance of a classification model at all classification thresholds. True ROC can only be computed for two-options, but we can consider each classification, i.e. prediction, correct or incorrect and overlay the curves. Note this means the lines may cover each other and be difficult to see.

Usage

computePseudoROCCurves(trueOutcomes, modelPercents)

Arguments

Details

This function requires the package 'pROC' to be installed.

Value

ggplot object containing the ROC curves.

Examples

percents <- data.frame(c(0.980, 0.675, 0.878, 0.303, 0.457, 0.758,
                         0.272, 0.524, 0.604, 0.342, 0.214, 0.569,
                         0.279, 0.128, 0.462, 0.098, 0.001, 0.187),
                       c(0.005, 0.160, 0.100, 0.244, 0.174, 0.143,
                         0.652, 0.292, 0.040, 0.312, 0.452, 0.168,
                         0.173, 0.221, 0.281, 0.029, 0.005, 0.057),
                       c(0.015, 0.165, 0.022, 0.453, 0.369, 0.099,
                         0.076, 0.084, 0.156, 0.346, 0.334, 0.263,
                         0.548, 0.651, 0.257, 0.873, 0.994, 0.756))
colnames(percents) <- c('0','1','2')
proc <- computePseudoROCCurves(c(0, 0, 0, 0, 0, 0,
                                 1, 1, 1, 1, 1, 1,
                                 2, 2, 2, 2, 2, 2),
                               percents)

Compute a Modified Random Forest Model

Description

This function creates a modified random forest model for principal component and meta-data. This can be useful to get a final model, but we recommend use of randomForest_CVPC in general, which includes the final model.

Usage

funkyForest(
  data,
  outcome = colnames(data)[1],
  unit = colnames(data)[2],
  nTrees = 500,
  varImpPlot = TRUE,
  metaNames = NULL,
  keepModels = TRUE,
  varSelPercent = 0.8,
  method = "class"
)

Arguments

Value

A list with entries

1. varImportanceData: Data.frame for variable importance information.

2. (Optional) model: List of CART that builds the random forest model.

3. (Optional) varImportancePlot: Variable importance plots.

Examples

ff <- funkyForest(
  data = TNBC[, c(1:8, ncol(TNBC))],
  outcome = "Class", unit = "Person",
  metaNames = c("Age")
  )

Fit a Modified Random Forest Model with Bounds and Alignment

Description

The function fits a modified random forest model to principal components of spatial interactions as well as meta-data. Additionally permutation and cross-validation is employed to improve understanding of the data.

Usage

funkyModel(
  data,
  K = 10,
  outcome = colnames(data)[1],
  unit = colnames(data)[2],
  metaNames = NULL,
  synthetics = 100,
  alpha = 0.05,
  silent = FALSE,
  rGuessSims = 500,
  subsetPlotSize = 25,
  nTrees = 500,
  method = "class"
)

Arguments

Value

List with the following items:

1. model: The funkyForest Model fit on the entire given data.

2. VariableImportance: Data.frame with the results of variable importance indices from the models and CV. The columns are var, est, sd, and cvSD.

3. AccuracyEstimate: Data.frame with model accuracy estimates: out-of-bag accuracy (OOB), biased estimate (bias), and random guess (guess). The columns are OOB, bias, and guess.

4. NoiseCutoff: Numeric indicating noise cutoff (vertical line).

5. InterpolationCutoff: Vector of numerics indicating the interpolation cutoff (curved line).

6. AdditionalParams: List of additional parameters for reference: Alpha and subsetPlotSize.

7. viPlot: ggplot2 object for vi plot with standardized results. It displays ordered underlying functions and meta-variables with point estimates, sd, noise cutoff, and interpolation cutoff all based on variable importance values.

8. subset_viPlot: (Optional) ggplot2 object for vi plot with standardized results and only top subsetPlotSize variables. It displays ordered underlying functions and meta-variables with point estimates, sd, noise cutoff, and interpolation cutoff all based on variable importance values.

Examples

# Parameters are reduced beyond recommended levels for speed
fm <- funkyModel(
  data = TNBC[, c(1:8, ncol(TNBC))],
  outcome = "Class", unit = "Person",
  metaNames = c("Age"),
  nTrees = 5, synthetics = 10,
  silent = TRUE
)

Get Agent Count Data

Description

This function gets the average percent agent counts per replicate, if there are replicates (i.e. replicate is not NULL), then the agent percents are calculated for each replicate and these percentages are averaged.

Usage

getCountData(
  agent_data,
  outcome,
  unit,
  replicate = NULL,
  type = "type",
  data_append = NULL
)

Arguments

Value

List with two elements:

• dat: Data.frame with outcome, unit, data_append, and the count data. Columns of the count data are named after the type and are given in the next list entry.

• agents: Vector of the the types, i.e. the column names for the new count data. This can be treated as meta data for funkyForest.

Examples

data_ct <- getCountData(TNBC_pheno[TNBC_pheno$Phenotype %in% c('Tumor','B'),],
                        outcome="Class", unit="Person",type="Phenotype")

Get K function

Description

This function computes the K function between the two agents for each unit, potentially averaging over replicates, or repeated measures.

Usage

getKFunction(
  data,
  agents,
  unit,
  replicate = NULL,
  rCheckVals = NULL,
  xRange = NULL,
  yRange = NULL,
  edgeCorrection = "isotropic"
)

Arguments

Value

data.frame with the first column being the checked radius and the remaining columns relating to the K function for each unit at those points. If a K function could not be computed, perhaps due to lack of data, an NA is returned for the K function.

Examples

KFunction <- getKFunction(
  agents = c("B", "Tumor"), unit = "Person",
  data = TNBC_pheno[TNBC_pheno$Person == 1, -1],
  rCheckVals = seq(0, 50, 1),
  edgeCorrection = "isotropic"
)

Get K Functions and Compute Principal Components

Description

This function computes K functions from point process data then converts it into PCs. Note, if there are replicates, i.e. multiple observations per unit, the K functions will be a weighted average based on the number of the first agents.

Usage

getKsPCAData(
  data,
  outcome = colnames(data)[1],
  unit = colnames(data)[5],
  replicate = NULL,
  rCheckVals = NULL,
  nPCs = 3,
  agents_df = as.data.frame(expand.grid(unique(data[, 4]), unique(data[, 4]))),
  xRange = NULL,
  yRange = NULL,
  edgeCorrection = "isotropic",
  nbasis = 21,
  silent = FALSE,
  displayTVE = FALSE
)

Arguments

Value

Data.frame with the outcome, unit and principle components of computed K functions.

Examples

dataPCA_pheno <- getKsPCAData(
  data = TNBC_pheno, unit = "Person",
  agents_df = data.frame(rep("B", 2), c("Tumor", "Fake")),
  nPCs = 3,
  rCheckVals = seq(0, 50, 1),
  displayTVE = TRUE
)

Plot Spatial Point Process

Description

This function is used to plot a spatial point process. This does not split data and instead puts all given data on a single plot.

Usage

plotPP(
  data,
  colorGuide = NULL,
  ptSize = 1,
  xlim = c(min(data[, 1]), max(data[, 1])),
  ylim = c(min(data[, 2]), max(data[, 2])),
  dropAxes = FALSE,
  layerBasedOnFrequency = TRUE,
  colors = NULL
)

Arguments

Value

ggplot2 plot of the spatial point process.

Examples

ppplot <- plotPP(
  TNBC_pheno[
    TNBC_pheno$Person == 1,
    c("cellx", "celly", "Phenotype")
  ],
  colorGuide = "none"
)

Compare K Functions Between outcomes

Description

This function plots K functions from different outcomes for comparison. Group means are included as bold lines. Additionally a reference line for a spatially random process can be included.

Usage

plot_K_functions(data, inc.legend = TRUE, inc.noise = FALSE)

Arguments

Value

ggplot2 object showing the K function with a superimposed average.

Examples

# Example 1
tmp <- getKFunction(TNBC_pheno[TNBC_pheno$Class == 0, -1],
  agents = c("Tumor", "Tumor"),
  unit = "Person",
  rCheckVals = seq(0, 50, 1)
)
tmp1 <- getKFunction(TNBC_pheno[TNBC_pheno$Class == 1, -1],
  agents = c("Tumor", "Tumor"),
  unit = "Person",
  rCheckVals = seq(0, 50, 1)
)
tmp_1 <- tidyr::pivot_longer(data = tmp, cols = K1:K18)
tmp1_1 <- tidyr::pivot_longer(data = tmp1, cols = K1:K15)

data_plot <- rbind(
  data.frame(
    "r" = tmp_1$r,
    "K" = tmp_1$value,
    "unit" = tmp_1$name,
    "outcome" = "0"
  ),
  data.frame(
    "r" = tmp1_1$r,
    "K" = tmp1_1$value,
    "unit" = paste0(tmp1_1$name, "_1"),
    "outcome" = "1"
  )
)

pk1 <- plot_K_functions(data_plot)

# Example 2
tmp <- getKFunction(TNBC_pheno[TNBC_pheno$Class == 0, -1],
  agents = c("Tumor", "B"), unit = "Person",
  rCheckVals = seq(0, 50, 1)
)
tmp1 <- getKFunction(TNBC_pheno[TNBC_pheno$Class == 1, -1],
  agents = c("Tumor", "B"), unit = "Person",
  rCheckVals = seq(0, 50, 1)
)

tmp_1 <- tidyr::pivot_longer(data = tmp, cols = K1:K18)
tmp1_1 <- tidyr::pivot_longer(data = tmp1, cols = K1:K15)

data_plot <- rbind(
  data.frame(
    "r" = tmp_1$r,
    "K" = tmp_1$value,
    "unit" = tmp_1$name,
    "outcome" = "0"
  ),
  data.frame(
    "r" = tmp1_1$r,
    "K" = tmp1_1$value,
    "unit" = paste0(tmp1_1$name, "_1"),
    "outcome" = "1"
  )
)

pk2 <- plot_K_functions(data_plot)

Predict a funkyForest

Description

This function gets the predicted value from a funkyForest model.

Usage

predict_funkyForest(model, data_pred, type = "all", data = NULL)

Arguments

Value

The returned data depends on type:

• type='pred': returns a vector of the predictions

• type='all': returns a vector of the predictions

Examples

data_pp <- simulatePP(
  agentVarData =
    data.frame(
      "outcome" = c(0, 1),
      "A" = c(0, 0),
      "B" = c(1 / 50, 1 / 50)
    ),
  agentKappaData = data.frame(
    "agent" = c("A", "B"),
    "clusterAgent" = c(NA, "A"),
    "kappa" = c(10, 5)
  ),
  unitsPerOutcome = 5,
  replicatesPerUnit = 1,
  silent = FALSE
)
pcaData <- getKsPCAData(data_pp,
  replicate = "replicate",
  xRange = c(0, 1), yRange = c(0, 1), silent = FALSE
)
RF <- funkyForest(data = pcaData[-2], nTrees = 5) #
pred <- predict_funkyForest(
  model = RF$model, type = "all",
  data_pred = pcaData[-2],
  data = pcaData[-2]
)

Simulate Meta Variables

Description

This function simulates meta-variables with varying distributions to append to some data.

Usage

simulateMeta(
  data,
  outcome = colnames(data)[1],
  metaInfo = data.frame(var = c("randUnif", "randBin", "rNorm", "corrUnif", "corrBin",
    "corrNorm"), rdist = c("runif", "rbinom", "rnorm", "runif", "rbinom", "rnorm"),
    outcome_0 = c("0.5", "0.5", "1", "0.5", "0.6", "1"), outcome_1 = c("0.5", "0.5", "1",
    "0.75", "0.65", "1.5"), outcome_2 = c("0.5", "0.5", "1", "0.95", "0.75", "1.5"))
)

Arguments

Details

Notes: runif may induce useless information so don't recommend correlating it

Value

Data.frame of the original data with meta-variables appended (as columns) at the end.

Examples

data <- simulatePP(
  agentVarData = data.frame(
    "outcome" = c(0, 1, 2),
    "A" = c(0, 0, 0),
    "B" = c(1 / 100, 1 / 500, 1 / 1000)
  ),
  agentKappaData = data.frame(
    "agent" = c("A", "B"),
    "clusterAgent" = c(NA, "A"),
    "kappa" = c(10, 3)
  ),
  unitsPerOutcome = 5,
  replicatesPerUnit = 1
)
pcaData <- getKsPCAData(
  data = data, replicate = "replicate",
  xRange = c(0, 1), yRange = c(0, 1)
)
pcaMeta <- simulateMeta(pcaData)

## Another simple example
data <- simulateMeta(
  data.frame("outcome" = c(0, 0, 0, 1, 1, 1), "unit" = 1:6)
)

Simulate a Point Process

Description

This function simulates a point pattern with optional clustering (visible and invisible). Multiple outcomes, units, and replicates are possible, e.g. a 3 stage disease (outcomes) over 20 people (units) with 3 images each (replicates).

Usage

simulatePP(
  agentVarData = data.frame(outcome = c(0, 1, 2), A = c(0, 0, 0), B = c(1/100, 1/500,
    1/500), C = c(1/500, 1/250, 1/100), D = c(1/100, 1/100, 1/100), E = c(1/500, 1/500,
    1/500), F = c(1/250, 1/250, 1/250)),
  agentKappaData = data.frame(agent = c("A", "B", "C", "D", "E", "F"), clusterAgent =
    c(NA, "A", "B", "C", NA, "A"), kappa = c(20, 5, 4, 2, 15, 5)),
  unitsPerOutcome = 20,
  replicatesPerUnit = 5,
  silent = FALSE
)

Arguments

Value

Data.frame containing each point the defined patterns.

The data.frame has columns for outcome, x coordinate, y coordinate, agent type, unit, and replicate id.

Examples

data <- simulatePP(
  agentVarData = data.frame(
    "outcome" = c(0, 1),
    "A" = c(0, 0),
    "B" = c(1 / 100, 1 / 500),
    "C" = c(1 / 500, 1 / 250),
    "D" = c(1 / 100, 1 / 100),
    "E" = c(1 / 500, 1 / 500)
  ),
  agentKappaData = data.frame(
    "agent" = c("A", "B", "C", "D", "E"),
    "clusterAgent" = c(NA, "A", "B", "C", NA),
    "kappa" = c(10, 3, 2, 1, 8)
  ),
  unitsPerOutcome = 4,
  replicatesPerUnit = 1
)