The kdca package implements a statistical test to assess whether a pathway is differentially co-expressed. The inputs into kdca are pathway expression values and a risk factor(s) of interest. The package can model multiple risk factors, covariates, and/or batch effects. kdca also accounts for variance-specific effects from the risk factors (or other variables) to avoid false discoveries. The risk factors can be continuous, categorical, or a mixture of both. Finally, we assume an appropriate normalization has been applied to the expression values (e.g., logCPM). For RNA-seq data, we also recommend treating library size as a covariate in the model (see below).

kdca captures complex differential co-expression using the linear, projection, and Gaussian kernels. Because pathway architecture is unknown, it is difficult to select the optimal kernel. As such, we provide an aggregate version (`Combined') that maximizes power across kernels. In future updates, users will be able to add their own kernel functions.

The methods implemented in this package are described in:

Bass AJ, Cutler DJ, Epstein MP. A powerful framework for differential co-expression analysis of continuous and categorical risk factors. Submitted; 2024.

To report any bugs or issues related to usage please report it on GitHub at https://github.com/ajbass/kdca.

To install the development version of the package:

# install development version of package
install.packages("devtools")
library("devtools")
devtools::install_github("ajbass/kdca")

We demonstrate the functionality of the kdca package using simulated data with no differential co-expression. First, load the kdca package and simulate data:

library(kdca)
# set seed
set.seed(123)

# Generate a categorical risk factor (sample size 100)
x <- matrix(rbinom(100, size = 1, prob = 0.5), ncol = 1)

# Generate a continuous covariate
z <-  matrix(rnorm(100 * 1), ncol = 1)

# Generate a pathway with 10 genes (no differential co-expression)
# with a variance-specific effect
ve <- rep(x + 1, 5)
error <- matrix(rnorm(100 * 10, sd = ve), ncol = 10)
beta <- matrix(1, ncol = 10) # effect sizes
y <- z %*% beta + x %*% beta + error # pathway

We generated a risk factor x that is categorical, a covariate z that is continuous, and a pathway with 10 genes (100 samples). The risk factor and covariate impact the expression values but there is no differential co-expression. Additionally, the risk factor has a strong variance-specific effect. We can then run the kdca function on this data set:

kdca_out <- kdca(x, y, mean_adjust = z, type_x = "categorical")

Note that the argument type_x specifies the risk factor type and the covariates/batch effects can be inputted into the argument mean_adjust.

We can apply kdca to conituous risk factors as well. Let us now suppose that z is the risk factor of interest and x is a covariate:

kdca_out <- kdca(z, y, mean_adjust = x, type_x = "continuous")

Finally, if x and z are risk factors of interest, then we can apply kdca to multiple risk factors:

kdca_out <- kdca(cbind(x,z), y, type_x = "continuous")

When inputting multiple risk factors, you have to set type_x = "continuous". See ?kdca for additional arguments.