Estimation

This Vignette covers the estimation function in depth.

The main component of dineR is the estimation function that enables any user the ability to perform differential network estimation with great ease. The need for such a function stems for the fact that in high dimensional settings, the inverse of any matrix cannot be determined analytical. One such avenue in which this is overcome is through the use of estimation procedures.

dineR includes the ability to perform this estimation under a number of different loss functions available within literature as well as the option to fine tune each of the estimation parameters to improve the speed and accuracy of the process given specific problem characteristics.

The focus of dineR lies primarily on differential networks, and the subsequent estimation thereof. However, the structure of data suitable for differential network estimation may not be immediately apparent. As such, this function provides users with an easy, scalable manner to generate appropriate data as they familiarize themselves with the estimation options available within dineR

Getting Started

The data_generator function enables users to generate two multivariate normal samples of varying sizes, whose number of features/dimensions are equal. These two samples are representative of the two experimental conditions under which one would typically determine a differential network.

Due to the technicalities of differential network estimation, it is not possible to validate the accuracy of estimates obtained through the use of traditional statistical methodologies. data_generator aims to address this by generating data whose covariance matrices, and precision matrices are known a-priori and as such so too is the differential network. This allows users to validate the estimation procedure for their specific experimental criteria.

Arguments

The most important arguments available within data_generator are as follows:

# Number of observations in sample 1
n_X <- 100 

# Number of observations in sample 2
n_Y <- 150

# Number of features in each of the samples
p <- 50

# The form of the precision covariance matrices
case <- "sparse"

# The seed of the simulation process to ensure reproducibility of results
seed <- 123

Data Generation

Having defined each of the key arguments above, the data_generator function can now be called:

data <- data_generator(n_X = n_X, n_Y = n_Y, p = p, case = case, seed = seed)

Outputs

Given the data has now been generated, the data can be “wrangled” for use within the estimation function.

# Extract the first sample
X <- data$X
cat("The number of observations in the first sample is:", nrow(X))
#> The number of observations in the first sample is: 100
cat("The number of features/dimensions in the first sample is:", ncol(X))
#> The number of features/dimensions in the first sample is: 50

# Extract the second sample
Y <- data$Y
cat("The number of observations in the second sample is:", nrow(Y))
#> The number of observations in the second sample is: 150
cat("The number of features/dimensions in the second sample is:", ncol(Y))
#> The number of features/dimensions in the second sample is: 50

It is also possible to extract the sample covariance matrices for each of the samples:

# Extract the first sample's covariance matrix
Sigma_X <- data$Sigma_X

# Extract the second sample's covariance matrix
Sigma_Y <- data$Sigma_Y

In addition to the above matrices, the precision matrices can also be extracted:

# Extract the first sample's precision matrix
Omega_X <- data$Omega_X

# Extract the second sample's precision matrix
Omega_Y <- data$Omega_Y

Lastly, it is possible to extract the differential network for comparison to the estimate:

# Extract the differential network
Delta <- data$Delta

Having covered each of the key arguments and outputs available within data_generator, please kindly note that further details are available within the function documentation as well as the listed reference.