marbles

MultivARiate Bayesian Layered variablE Selection

R package for "Mohammed, S., Kurtek, S., Bharath, K., Rao, A. and Baladandayuthapani, V., 2021. Tumor Radiogenomics with Bayesian Layered Variable Selection."

Code to run the model in (Mohammed et.al, 2021) on the imaging and genomic data in gliomas using the package marbles.

Contents

1. Accessing low grade glioma (LGG) data from the package
2. Fit the multivariate Bayesian layered variable selection model for LGG data
3. Plot radiogenomic associations in LGG

Accessing low grade glioma data from the package

Installing the package

# install the package (devtools package needed)
if(!require(devtools)) install.packages("devtools")
devtools::install_github('shariq-mohammed/marbles')

Load the package

library(marbles)

Example

The complete description for each of the data files can be accessed in R or RStudio as follows:

?cancer_gene_exp_LGG
?mri_data
?pdfs
?pc_scores
?group_id

Data can be accessed directly using the variable names once the package is loaded in the environment. For example, the structure and dimensions of the cancer_gene_exp_LGG object can be checked as follows

# structure
str(cancer_gene_exp_LGG)

# dimensions
dim(cancer_gene_exp_LGG)

Fit the multivariate Bayesian layered variable selection model for LGG data

Run the sequential estimation model for different values of $v_0$.

# sequence of values for v_0
v0.seq = seq(from = 0.001, by = 0.001, length.out = 40)

Estimate the model parameters with pc_scores as the multivariate response and cancer_gene_exp_LGG as the predictors.

res = seq_est_model_selection(Y = pc_scores, X = cancer_gene_exp_LGG,
                              g_id = group_id, nCores = 4, v0seq = v0.seq)

Plot radiogenomic associations in LGG

Create a disc plot using MR imaging sequence names and gene names.

# imaging sequence names
im.seq = c('FLAIR', 'T1', 'T1Gd', 'T2')

# gene names
X.colnames = colnames(cancer_gene_exp_LGG)

disc_plot(res$seq_res$w, im.seq, X.colnames, layer = T)