dineR

Overview

dineR is a R package, that aims to enable users of all backgrounds to easily, and computationally efficiently perform differential network estimation. Data can either be provided directly, or simulated. The differential network is then efficiently estimated through the use of a selected loss function and either sequential or parallel optimization.

Installation

You can install the latest release version of dineR from CRAN with:

install.packages("dineR")

However, it is also possible to install the development version from GitHub with:

devtools::install_github("RicSalgado/dineR")

Usage

A basic workflow of estimating a differential network using dineR is as follows:

# Load dineR into the current R session
library(dineR)

Data Generation

dineR allows for any qualifying data to be used to estimate a differential network.

For the purposes of this demonstration, we will simulate data using dineR’s in-house functions.

# The sample size of the first sample:
n_X <- 100

# The sample size of the second sample:
n_Y <- n_X

# The number of features within our first sample:
p_X <- 100

# The number of features within our second sample:
p_Y <- p_X

# The covariance matrix structure of interest:
case <- "sparse"

# Generating the data:
data <- data_generator(n = n_X, p = p_X, seed = 123)

# Extracting the relevent samples:
X <- data$X
Y <- data$Y

An advantage of using the above approach to simulate data, over other options within in R is that the data is guaranteed to be of the form required by dineR, but it also produces the inverse covariance matrices and as such it is possible to evaluate the accuracy of the estimation approach.

To evaluate the analytical differential network simply call the following:

diff_Omega <- data$diff_Omega

Estimation

To perform differential network estimation, there are a variety of options provided to the user to fit every use case and scenario. Each of these options comes with an appropriately selected default allowing users to get up and running with their estimation with greater ease.

The most basic function call in which only the two samples are specified would then be as follows:

results <- estimation(X, Y)

For further details regarding each of the available options, the reader is encouraged to review the function documentation and accompanying literature.

Outputs

Having completed the differential network estimation, there are a number of outputs available to view to confirm the estimation has behaved as anticipated.

The output of greatest interest however is easily the estimated differential network which can be accessed as follows:

First_Estimate <- result$path[[1]][1:5, 1:5]

The above code extracts the first 5 rows, and 5 columns of the differential network estimate obtained for the first value of the tuning parameter $\lambda$.

Getting more involved

The above example is an extremely high-level introduction into the topic of differential network estimation through the use of dineR and it is recommended any users who wish to explore the topic further consult the available documentation for the relevant details.

Happy estimating.