In JacobHelwig/covdepGE: Covariate Dependent Graph Estimation

Here we demonstrate an application of covdepGE to real-world data similar to that which was used in (1). In both cases, the protein expression values for genes from patients with Breast Invasive Carcinoma (BRCA) are modeled, with data sourced from The Cancer Genome Atlas (TCGA). (1) considers the following genes :

• CTNNB1
• BRCA2
• MET
• E-cadherin
• N-cadherin
• NFkB1
• snail
• STAT3

Here, we consider the genes:

• STAT3_pY705
• E-Cadherin
• N-Cadherin
• NF-kB-p65_pS536

While (1) models the precision matrix for these genes as a function of the FOXC2 gene, here we use FOXO3a. After fitting the model, we visualize each of the estimated graphs (left column) alongside the corresponding FOXO3a values (right column).

# BiocManager::install("RTCGA.BRCA")
library(covdepGE)
library(ggpubr)

# load data
dat <- RTCGA.RPPA::BRCA.RPPA
zvar <- "FOXO3a"
Z <- dat[ , zvar]
vars <- c("STAT3_pY705",
          "E-Cadherin",
          "N-Cadherin",
          "NF-kB-p65_pS536")
X <- dat[ , vars]

# fit model
out <- covdepGE(X, Z, parallel = T, num_workers = 4)
out

# returns graph visualization and histogram
plot_brca <- function(graph, z, xlim, ylim, color, labs, labz, bw){
  graph_viz <- matViz(graph, color2 = color) + 
    theme(legend.position='none', axis.text.x = element_text(angle=30,hjust=1,vjust=1.0)) + 
    scale_y_continuous(labels=labs, breaks=1:nrow(graph)) + 
    scale_x_continuous(labels=labs, breaks=1:nrow(graph))
  hist <- ggplot() + geom_histogram(aes(z), color='black', fill = color, binwidth = bw) + 
    xlab(labz) + theme_bw() + xlim(xlim) + ylim(c(0, ylim))
  list(graph_viz, hist)
}

# get graphs and corresponding observations
graphs <- lapply(out$graphs$unique_graphs, `[[`, "graph")
inds <- lapply(out$graphs$unique_graphs, `[[`, "indices")

# sort graphs by minimum Z value
ord <- order(sapply(inds, function(i) min(Z[i])))
colors <- ggsci::pal_nejm('default')(length(ord))

# plot results
get_graphs <- function(i){
  plot_brca(graph=graphs[[i]], z=Z[inds[[i]]], xlim=range(Z), ylim=25, color=colors[i], 
            labs=vars, labz=zvar, bw=0.04)
}
plots_list <- Reduce(c, lapply(ord, get_graphs))
ggarrange(plotlist = plots_list, ncol=2, nrow=length(ord))

Bibliography

(1) Dasgupta, Sutanoy, Peng Zhao, Jacob Helwig, Prasenjit Ghosh, Debdeep Pati, and Bani K. Mallick. "An Approximate Bayesian Approach to Covariate-dependent Graphical Modeling." arXiv preprint arXiv:2303.08979 (2023).

JacobHelwig/covdepGE documentation built on April 11, 2024, 7:22 a.m.